1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * derived from drivers/kvm/kvm_main.c
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright (C) 2008 Qumranet, Inc.
9 * Copyright IBM Corporation, 2008
10 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
13 * Avi Kivity <avi@qumranet.com>
14 * Yaniv Kamay <yaniv@qumranet.com>
15 * Amit Shah <amit.shah@qumranet.com>
16 * Ben-Ami Yassour <benami@il.ibm.com>
19 #include <linux/kvm_host.h>
25 #include "kvm_cache_regs.h"
26 #include "kvm_emulate.h"
34 #include <linux/clocksource.h>
35 #include <linux/interrupt.h>
36 #include <linux/kvm.h>
38 #include <linux/vmalloc.h>
39 #include <linux/export.h>
40 #include <linux/moduleparam.h>
41 #include <linux/mman.h>
42 #include <linux/highmem.h>
43 #include <linux/iommu.h>
44 #include <linux/intel-iommu.h>
45 #include <linux/cpufreq.h>
46 #include <linux/user-return-notifier.h>
47 #include <linux/srcu.h>
48 #include <linux/slab.h>
49 #include <linux/perf_event.h>
50 #include <linux/uaccess.h>
51 #include <linux/hash.h>
52 #include <linux/pci.h>
53 #include <linux/timekeeper_internal.h>
54 #include <linux/pvclock_gtod.h>
55 #include <linux/kvm_irqfd.h>
56 #include <linux/irqbypass.h>
57 #include <linux/sched/stat.h>
58 #include <linux/sched/isolation.h>
59 #include <linux/mem_encrypt.h>
60 #include <linux/entry-kvm.h>
61 #include <linux/suspend.h>
63 #include <trace/events/kvm.h>
65 #include <asm/debugreg.h>
70 #include <linux/kernel_stat.h>
71 #include <asm/fpu/internal.h> /* Ugh! */
72 #include <asm/pvclock.h>
73 #include <asm/div64.h>
74 #include <asm/irq_remapping.h>
75 #include <asm/mshyperv.h>
76 #include <asm/hypervisor.h>
77 #include <asm/tlbflush.h>
78 #include <asm/intel_pt.h>
79 #include <asm/emulate_prefix.h>
81 #include <clocksource/hyperv_timer.h>
83 #define CREATE_TRACE_POINTS
86 #define MAX_IO_MSRS 256
87 #define KVM_MAX_MCE_BANKS 32
88 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
89 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
91 #define emul_to_vcpu(ctxt) \
92 ((struct kvm_vcpu *)(ctxt)->vcpu)
95 * - enable syscall per default because its emulated by KVM
96 * - enable LME and LMA per default on 64 bit KVM
100 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
102 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
105 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
107 #define KVM_EXIT_HYPERCALL_VALID_MASK (1 << KVM_HC_MAP_GPA_RANGE)
109 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
110 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
112 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
113 static void process_nmi(struct kvm_vcpu *vcpu);
114 static void process_smi(struct kvm_vcpu *vcpu);
115 static void enter_smm(struct kvm_vcpu *vcpu);
116 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
117 static void store_regs(struct kvm_vcpu *vcpu);
118 static int sync_regs(struct kvm_vcpu *vcpu);
120 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
121 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
123 struct kvm_x86_ops kvm_x86_ops __read_mostly;
124 EXPORT_SYMBOL_GPL(kvm_x86_ops);
126 #define KVM_X86_OP(func) \
127 DEFINE_STATIC_CALL_NULL(kvm_x86_##func, \
128 *(((struct kvm_x86_ops *)0)->func));
129 #define KVM_X86_OP_NULL KVM_X86_OP
130 #include <asm/kvm-x86-ops.h>
131 EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits);
132 EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg);
133 EXPORT_STATIC_CALL_GPL(kvm_x86_tlb_flush_current);
135 static bool __read_mostly ignore_msrs = 0;
136 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
138 bool __read_mostly report_ignored_msrs = true;
139 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
140 EXPORT_SYMBOL_GPL(report_ignored_msrs);
142 unsigned int min_timer_period_us = 200;
143 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
145 static bool __read_mostly kvmclock_periodic_sync = true;
146 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
148 bool __read_mostly kvm_has_tsc_control;
149 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
150 u32 __read_mostly kvm_max_guest_tsc_khz;
151 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
152 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
153 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
154 u64 __read_mostly kvm_max_tsc_scaling_ratio;
155 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
156 u64 __read_mostly kvm_default_tsc_scaling_ratio;
157 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
158 bool __read_mostly kvm_has_bus_lock_exit;
159 EXPORT_SYMBOL_GPL(kvm_has_bus_lock_exit);
161 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
162 static u32 __read_mostly tsc_tolerance_ppm = 250;
163 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
166 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
167 * adaptive tuning starting from default advancement of 1000ns. '0' disables
168 * advancement entirely. Any other value is used as-is and disables adaptive
169 * tuning, i.e. allows privileged userspace to set an exact advancement time.
171 static int __read_mostly lapic_timer_advance_ns = -1;
172 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
174 static bool __read_mostly vector_hashing = true;
175 module_param(vector_hashing, bool, S_IRUGO);
177 bool __read_mostly enable_vmware_backdoor = false;
178 module_param(enable_vmware_backdoor, bool, S_IRUGO);
179 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
181 static bool __read_mostly force_emulation_prefix = false;
182 module_param(force_emulation_prefix, bool, S_IRUGO);
184 int __read_mostly pi_inject_timer = -1;
185 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
188 * Restoring the host value for MSRs that are only consumed when running in
189 * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
190 * returns to userspace, i.e. the kernel can run with the guest's value.
192 #define KVM_MAX_NR_USER_RETURN_MSRS 16
194 struct kvm_user_return_msrs {
195 struct user_return_notifier urn;
197 struct kvm_user_return_msr_values {
200 } values[KVM_MAX_NR_USER_RETURN_MSRS];
203 u32 __read_mostly kvm_nr_uret_msrs;
204 EXPORT_SYMBOL_GPL(kvm_nr_uret_msrs);
205 static u32 __read_mostly kvm_uret_msrs_list[KVM_MAX_NR_USER_RETURN_MSRS];
206 static struct kvm_user_return_msrs __percpu *user_return_msrs;
208 #define KVM_SUPPORTED_XCR0 (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
209 | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
210 | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
211 | XFEATURE_MASK_PKRU)
213 u64 __read_mostly host_efer;
214 EXPORT_SYMBOL_GPL(host_efer);
216 bool __read_mostly allow_smaller_maxphyaddr = 0;
217 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
219 bool __read_mostly enable_apicv = true;
220 EXPORT_SYMBOL_GPL(enable_apicv);
222 u64 __read_mostly host_xss;
223 EXPORT_SYMBOL_GPL(host_xss);
224 u64 __read_mostly supported_xss;
225 EXPORT_SYMBOL_GPL(supported_xss);
227 const struct _kvm_stats_desc kvm_vm_stats_desc[] = {
228 KVM_GENERIC_VM_STATS(),
229 STATS_DESC_COUNTER(VM, mmu_shadow_zapped),
230 STATS_DESC_COUNTER(VM, mmu_pte_write),
231 STATS_DESC_COUNTER(VM, mmu_pde_zapped),
232 STATS_DESC_COUNTER(VM, mmu_flooded),
233 STATS_DESC_COUNTER(VM, mmu_recycled),
234 STATS_DESC_COUNTER(VM, mmu_cache_miss),
235 STATS_DESC_ICOUNTER(VM, mmu_unsync),
236 STATS_DESC_ICOUNTER(VM, pages_4k),
237 STATS_DESC_ICOUNTER(VM, pages_2m),
238 STATS_DESC_ICOUNTER(VM, pages_1g),
239 STATS_DESC_ICOUNTER(VM, nx_lpage_splits),
240 STATS_DESC_PCOUNTER(VM, max_mmu_rmap_size),
241 STATS_DESC_PCOUNTER(VM, max_mmu_page_hash_collisions)
244 const struct kvm_stats_header kvm_vm_stats_header = {
245 .name_size = KVM_STATS_NAME_SIZE,
246 .num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
247 .id_offset = sizeof(struct kvm_stats_header),
248 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
249 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
250 sizeof(kvm_vm_stats_desc),
253 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
254 KVM_GENERIC_VCPU_STATS(),
255 STATS_DESC_COUNTER(VCPU, pf_fixed),
256 STATS_DESC_COUNTER(VCPU, pf_guest),
257 STATS_DESC_COUNTER(VCPU, tlb_flush),
258 STATS_DESC_COUNTER(VCPU, invlpg),
259 STATS_DESC_COUNTER(VCPU, exits),
260 STATS_DESC_COUNTER(VCPU, io_exits),
261 STATS_DESC_COUNTER(VCPU, mmio_exits),
262 STATS_DESC_COUNTER(VCPU, signal_exits),
263 STATS_DESC_COUNTER(VCPU, irq_window_exits),
264 STATS_DESC_COUNTER(VCPU, nmi_window_exits),
265 STATS_DESC_COUNTER(VCPU, l1d_flush),
266 STATS_DESC_COUNTER(VCPU, halt_exits),
267 STATS_DESC_COUNTER(VCPU, request_irq_exits),
268 STATS_DESC_COUNTER(VCPU, irq_exits),
269 STATS_DESC_COUNTER(VCPU, host_state_reload),
270 STATS_DESC_COUNTER(VCPU, fpu_reload),
271 STATS_DESC_COUNTER(VCPU, insn_emulation),
272 STATS_DESC_COUNTER(VCPU, insn_emulation_fail),
273 STATS_DESC_COUNTER(VCPU, hypercalls),
274 STATS_DESC_COUNTER(VCPU, irq_injections),
275 STATS_DESC_COUNTER(VCPU, nmi_injections),
276 STATS_DESC_COUNTER(VCPU, req_event),
277 STATS_DESC_COUNTER(VCPU, nested_run),
278 STATS_DESC_COUNTER(VCPU, directed_yield_attempted),
279 STATS_DESC_COUNTER(VCPU, directed_yield_successful),
280 STATS_DESC_ICOUNTER(VCPU, guest_mode)
283 const struct kvm_stats_header kvm_vcpu_stats_header = {
284 .name_size = KVM_STATS_NAME_SIZE,
285 .num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
286 .id_offset = sizeof(struct kvm_stats_header),
287 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
288 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
289 sizeof(kvm_vcpu_stats_desc),
292 u64 __read_mostly host_xcr0;
293 u64 __read_mostly supported_xcr0;
294 EXPORT_SYMBOL_GPL(supported_xcr0);
296 static struct kmem_cache *x86_fpu_cache;
298 static struct kmem_cache *x86_emulator_cache;
301 * When called, it means the previous get/set msr reached an invalid msr.
302 * Return true if we want to ignore/silent this failed msr access.
304 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
306 const char *op = write ? "wrmsr" : "rdmsr";
309 if (report_ignored_msrs)
310 kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
315 kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
321 static struct kmem_cache *kvm_alloc_emulator_cache(void)
323 unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
324 unsigned int size = sizeof(struct x86_emulate_ctxt);
326 return kmem_cache_create_usercopy("x86_emulator", size,
327 __alignof__(struct x86_emulate_ctxt),
328 SLAB_ACCOUNT, useroffset,
329 size - useroffset, NULL);
332 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
334 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
337 for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
338 vcpu->arch.apf.gfns[i] = ~0;
341 static void kvm_on_user_return(struct user_return_notifier *urn)
344 struct kvm_user_return_msrs *msrs
345 = container_of(urn, struct kvm_user_return_msrs, urn);
346 struct kvm_user_return_msr_values *values;
350 * Disabling irqs at this point since the following code could be
351 * interrupted and executed through kvm_arch_hardware_disable()
353 local_irq_save(flags);
354 if (msrs->registered) {
355 msrs->registered = false;
356 user_return_notifier_unregister(urn);
358 local_irq_restore(flags);
359 for (slot = 0; slot < kvm_nr_uret_msrs; ++slot) {
360 values = &msrs->values[slot];
361 if (values->host != values->curr) {
362 wrmsrl(kvm_uret_msrs_list[slot], values->host);
363 values->curr = values->host;
368 static int kvm_probe_user_return_msr(u32 msr)
374 ret = rdmsrl_safe(msr, &val);
377 ret = wrmsrl_safe(msr, val);
383 int kvm_add_user_return_msr(u32 msr)
385 BUG_ON(kvm_nr_uret_msrs >= KVM_MAX_NR_USER_RETURN_MSRS);
387 if (kvm_probe_user_return_msr(msr))
390 kvm_uret_msrs_list[kvm_nr_uret_msrs] = msr;
391 return kvm_nr_uret_msrs++;
393 EXPORT_SYMBOL_GPL(kvm_add_user_return_msr);
395 int kvm_find_user_return_msr(u32 msr)
399 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
400 if (kvm_uret_msrs_list[i] == msr)
405 EXPORT_SYMBOL_GPL(kvm_find_user_return_msr);
407 static void kvm_user_return_msr_cpu_online(void)
409 unsigned int cpu = smp_processor_id();
410 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
414 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
415 rdmsrl_safe(kvm_uret_msrs_list[i], &value);
416 msrs->values[i].host = value;
417 msrs->values[i].curr = value;
421 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
423 unsigned int cpu = smp_processor_id();
424 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
427 value = (value & mask) | (msrs->values[slot].host & ~mask);
428 if (value == msrs->values[slot].curr)
430 err = wrmsrl_safe(kvm_uret_msrs_list[slot], value);
434 msrs->values[slot].curr = value;
435 if (!msrs->registered) {
436 msrs->urn.on_user_return = kvm_on_user_return;
437 user_return_notifier_register(&msrs->urn);
438 msrs->registered = true;
442 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
444 static void drop_user_return_notifiers(void)
446 unsigned int cpu = smp_processor_id();
447 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
449 if (msrs->registered)
450 kvm_on_user_return(&msrs->urn);
453 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
455 return vcpu->arch.apic_base;
457 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
459 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
461 return kvm_apic_mode(kvm_get_apic_base(vcpu));
463 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
465 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
467 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
468 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
469 u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
470 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
472 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
474 if (!msr_info->host_initiated) {
475 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
477 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
481 kvm_lapic_set_base(vcpu, msr_info->data);
482 kvm_recalculate_apic_map(vcpu->kvm);
485 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
488 * Handle a fault on a hardware virtualization (VMX or SVM) instruction.
490 * Hardware virtualization extension instructions may fault if a reboot turns
491 * off virtualization while processes are running. Usually after catching the
492 * fault we just panic; during reboot instead the instruction is ignored.
494 noinstr void kvm_spurious_fault(void)
496 /* Fault while not rebooting. We want the trace. */
497 BUG_ON(!kvm_rebooting);
499 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
501 #define EXCPT_BENIGN 0
502 #define EXCPT_CONTRIBUTORY 1
505 static int exception_class(int vector)
515 return EXCPT_CONTRIBUTORY;
522 #define EXCPT_FAULT 0
524 #define EXCPT_ABORT 2
525 #define EXCPT_INTERRUPT 3
527 static int exception_type(int vector)
531 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
532 return EXCPT_INTERRUPT;
536 /* #DB is trap, as instruction watchpoints are handled elsewhere */
537 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
540 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
543 /* Reserved exceptions will result in fault */
547 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
549 unsigned nr = vcpu->arch.exception.nr;
550 bool has_payload = vcpu->arch.exception.has_payload;
551 unsigned long payload = vcpu->arch.exception.payload;
559 * "Certain debug exceptions may clear bit 0-3. The
560 * remaining contents of the DR6 register are never
561 * cleared by the processor".
563 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
565 * In order to reflect the #DB exception payload in guest
566 * dr6, three components need to be considered: active low
567 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
569 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
570 * In the target guest dr6:
571 * FIXED_1 bits should always be set.
572 * Active low bits should be cleared if 1-setting in payload.
573 * Active high bits should be set if 1-setting in payload.
575 * Note, the payload is compatible with the pending debug
576 * exceptions/exit qualification under VMX, that active_low bits
577 * are active high in payload.
578 * So they need to be flipped for DR6.
580 vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
581 vcpu->arch.dr6 |= payload;
582 vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
585 * The #DB payload is defined as compatible with the 'pending
586 * debug exceptions' field under VMX, not DR6. While bit 12 is
587 * defined in the 'pending debug exceptions' field (enabled
588 * breakpoint), it is reserved and must be zero in DR6.
590 vcpu->arch.dr6 &= ~BIT(12);
593 vcpu->arch.cr2 = payload;
597 vcpu->arch.exception.has_payload = false;
598 vcpu->arch.exception.payload = 0;
600 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
602 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
603 unsigned nr, bool has_error, u32 error_code,
604 bool has_payload, unsigned long payload, bool reinject)
609 kvm_make_request(KVM_REQ_EVENT, vcpu);
611 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
615 * On vmentry, vcpu->arch.exception.pending is only
616 * true if an event injection was blocked by
617 * nested_run_pending. In that case, however,
618 * vcpu_enter_guest requests an immediate exit,
619 * and the guest shouldn't proceed far enough to
622 WARN_ON_ONCE(vcpu->arch.exception.pending);
623 vcpu->arch.exception.injected = true;
624 if (WARN_ON_ONCE(has_payload)) {
626 * A reinjected event has already
627 * delivered its payload.
633 vcpu->arch.exception.pending = true;
634 vcpu->arch.exception.injected = false;
636 vcpu->arch.exception.has_error_code = has_error;
637 vcpu->arch.exception.nr = nr;
638 vcpu->arch.exception.error_code = error_code;
639 vcpu->arch.exception.has_payload = has_payload;
640 vcpu->arch.exception.payload = payload;
641 if (!is_guest_mode(vcpu))
642 kvm_deliver_exception_payload(vcpu);
646 /* to check exception */
647 prev_nr = vcpu->arch.exception.nr;
648 if (prev_nr == DF_VECTOR) {
649 /* triple fault -> shutdown */
650 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
653 class1 = exception_class(prev_nr);
654 class2 = exception_class(nr);
655 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
656 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
658 * Generate double fault per SDM Table 5-5. Set
659 * exception.pending = true so that the double fault
660 * can trigger a nested vmexit.
662 vcpu->arch.exception.pending = true;
663 vcpu->arch.exception.injected = false;
664 vcpu->arch.exception.has_error_code = true;
665 vcpu->arch.exception.nr = DF_VECTOR;
666 vcpu->arch.exception.error_code = 0;
667 vcpu->arch.exception.has_payload = false;
668 vcpu->arch.exception.payload = 0;
670 /* replace previous exception with a new one in a hope
671 that instruction re-execution will regenerate lost
676 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
678 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
680 EXPORT_SYMBOL_GPL(kvm_queue_exception);
682 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
684 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
686 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
688 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
689 unsigned long payload)
691 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
693 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
695 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
696 u32 error_code, unsigned long payload)
698 kvm_multiple_exception(vcpu, nr, true, error_code,
699 true, payload, false);
702 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
705 kvm_inject_gp(vcpu, 0);
707 return kvm_skip_emulated_instruction(vcpu);
711 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
713 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
715 ++vcpu->stat.pf_guest;
716 vcpu->arch.exception.nested_apf =
717 is_guest_mode(vcpu) && fault->async_page_fault;
718 if (vcpu->arch.exception.nested_apf) {
719 vcpu->arch.apf.nested_apf_token = fault->address;
720 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
722 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
726 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
728 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
729 struct x86_exception *fault)
731 struct kvm_mmu *fault_mmu;
732 WARN_ON_ONCE(fault->vector != PF_VECTOR);
734 fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
738 * Invalidate the TLB entry for the faulting address, if it exists,
739 * else the access will fault indefinitely (and to emulate hardware).
741 if ((fault->error_code & PFERR_PRESENT_MASK) &&
742 !(fault->error_code & PFERR_RSVD_MASK))
743 kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
744 fault_mmu->root_hpa);
746 fault_mmu->inject_page_fault(vcpu, fault);
747 return fault->nested_page_fault;
749 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
751 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
753 atomic_inc(&vcpu->arch.nmi_queued);
754 kvm_make_request(KVM_REQ_NMI, vcpu);
756 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
758 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
760 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
762 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
764 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
766 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
768 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
771 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
772 * a #GP and return false.
774 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
776 if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
778 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
781 EXPORT_SYMBOL_GPL(kvm_require_cpl);
783 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
785 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
788 kvm_queue_exception(vcpu, UD_VECTOR);
791 EXPORT_SYMBOL_GPL(kvm_require_dr);
793 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
795 return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
799 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
801 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
803 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
807 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
810 * If the MMU is nested, CR3 holds an L2 GPA and needs to be translated
813 real_gpa = mmu->translate_gpa(vcpu, gfn_to_gpa(pdpt_gfn),
814 PFERR_USER_MASK | PFERR_WRITE_MASK, NULL);
815 if (real_gpa == UNMAPPED_GVA)
818 /* Note the offset, PDPTRs are 32 byte aligned when using PAE paging. */
819 ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(real_gpa), pdpte,
820 cr3 & GENMASK(11, 5), sizeof(pdpte));
824 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
825 if ((pdpte[i] & PT_PRESENT_MASK) &&
826 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
831 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
832 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
833 vcpu->arch.pdptrs_from_userspace = false;
837 EXPORT_SYMBOL_GPL(load_pdptrs);
839 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
841 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
842 kvm_clear_async_pf_completion_queue(vcpu);
843 kvm_async_pf_hash_reset(vcpu);
846 if ((cr0 ^ old_cr0) & KVM_MMU_CR0_ROLE_BITS)
847 kvm_mmu_reset_context(vcpu);
849 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
850 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
851 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
852 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
854 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
856 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
858 unsigned long old_cr0 = kvm_read_cr0(vcpu);
859 unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
864 if (cr0 & 0xffffffff00000000UL)
868 cr0 &= ~CR0_RESERVED_BITS;
870 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
873 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
877 if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
878 (cr0 & X86_CR0_PG)) {
883 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
888 if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
889 is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
890 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
893 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
896 static_call(kvm_x86_set_cr0)(vcpu, cr0);
898 kvm_post_set_cr0(vcpu, old_cr0, cr0);
902 EXPORT_SYMBOL_GPL(kvm_set_cr0);
904 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
906 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
908 EXPORT_SYMBOL_GPL(kvm_lmsw);
910 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
912 if (vcpu->arch.guest_state_protected)
915 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
917 if (vcpu->arch.xcr0 != host_xcr0)
918 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
920 if (vcpu->arch.xsaves_enabled &&
921 vcpu->arch.ia32_xss != host_xss)
922 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
925 if (static_cpu_has(X86_FEATURE_PKU) &&
926 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
927 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
928 vcpu->arch.pkru != vcpu->arch.host_pkru)
929 write_pkru(vcpu->arch.pkru);
931 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
933 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
935 if (vcpu->arch.guest_state_protected)
938 if (static_cpu_has(X86_FEATURE_PKU) &&
939 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
940 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
941 vcpu->arch.pkru = rdpkru();
942 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
943 write_pkru(vcpu->arch.host_pkru);
946 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
948 if (vcpu->arch.xcr0 != host_xcr0)
949 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
951 if (vcpu->arch.xsaves_enabled &&
952 vcpu->arch.ia32_xss != host_xss)
953 wrmsrl(MSR_IA32_XSS, host_xss);
957 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
959 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
962 u64 old_xcr0 = vcpu->arch.xcr0;
965 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
966 if (index != XCR_XFEATURE_ENABLED_MASK)
968 if (!(xcr0 & XFEATURE_MASK_FP))
970 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
974 * Do not allow the guest to set bits that we do not support
975 * saving. However, xcr0 bit 0 is always set, even if the
976 * emulated CPU does not support XSAVE (see fx_init).
978 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
979 if (xcr0 & ~valid_bits)
982 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
983 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
986 if (xcr0 & XFEATURE_MASK_AVX512) {
987 if (!(xcr0 & XFEATURE_MASK_YMM))
989 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
992 vcpu->arch.xcr0 = xcr0;
994 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
995 kvm_update_cpuid_runtime(vcpu);
999 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
1001 if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
1002 __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
1003 kvm_inject_gp(vcpu, 0);
1007 return kvm_skip_emulated_instruction(vcpu);
1009 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1011 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1013 if (cr4 & cr4_reserved_bits)
1016 if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1019 return static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1021 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
1023 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1025 if (((cr4 ^ old_cr4) & KVM_MMU_CR4_ROLE_BITS) ||
1026 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1027 kvm_mmu_reset_context(vcpu);
1029 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1031 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1033 unsigned long old_cr4 = kvm_read_cr4(vcpu);
1034 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1037 if (!kvm_is_valid_cr4(vcpu, cr4))
1040 if (is_long_mode(vcpu)) {
1041 if (!(cr4 & X86_CR4_PAE))
1043 if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1045 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1046 && ((cr4 ^ old_cr4) & pdptr_bits)
1047 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1048 kvm_read_cr3(vcpu)))
1051 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1052 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1055 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1056 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1060 static_call(kvm_x86_set_cr4)(vcpu, cr4);
1062 kvm_post_set_cr4(vcpu, old_cr4, cr4);
1066 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1068 static void kvm_invalidate_pcid(struct kvm_vcpu *vcpu, unsigned long pcid)
1070 struct kvm_mmu *mmu = vcpu->arch.mmu;
1071 unsigned long roots_to_free = 0;
1075 * If neither the current CR3 nor any of the prev_roots use the given
1076 * PCID, then nothing needs to be done here because a resync will
1077 * happen anyway before switching to any other CR3.
1079 if (kvm_get_active_pcid(vcpu) == pcid) {
1080 kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
1081 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1084 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
1085 if (kvm_get_pcid(vcpu, mmu->prev_roots[i].pgd) == pcid)
1086 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
1088 kvm_mmu_free_roots(vcpu, mmu, roots_to_free);
1091 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1093 bool skip_tlb_flush = false;
1094 unsigned long pcid = 0;
1095 #ifdef CONFIG_X86_64
1096 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1099 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1100 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1101 pcid = cr3 & X86_CR3_PCID_MASK;
1105 /* PDPTRs are always reloaded for PAE paging. */
1106 if (cr3 == kvm_read_cr3(vcpu) && !is_pae_paging(vcpu))
1107 goto handle_tlb_flush;
1110 * Do not condition the GPA check on long mode, this helper is used to
1111 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1112 * the current vCPU mode is accurate.
1114 if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1117 if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1120 if (cr3 != kvm_read_cr3(vcpu))
1121 kvm_mmu_new_pgd(vcpu, cr3);
1123 vcpu->arch.cr3 = cr3;
1124 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1128 * A load of CR3 that flushes the TLB flushes only the current PCID,
1129 * even if PCID is disabled, in which case PCID=0 is flushed. It's a
1130 * moot point in the end because _disabling_ PCID will flush all PCIDs,
1131 * and it's impossible to use a non-zero PCID when PCID is disabled,
1132 * i.e. only PCID=0 can be relevant.
1134 if (!skip_tlb_flush)
1135 kvm_invalidate_pcid(vcpu, pcid);
1139 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1141 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1143 if (cr8 & CR8_RESERVED_BITS)
1145 if (lapic_in_kernel(vcpu))
1146 kvm_lapic_set_tpr(vcpu, cr8);
1148 vcpu->arch.cr8 = cr8;
1151 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1153 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1155 if (lapic_in_kernel(vcpu))
1156 return kvm_lapic_get_cr8(vcpu);
1158 return vcpu->arch.cr8;
1160 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1162 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1166 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1167 for (i = 0; i < KVM_NR_DB_REGS; i++)
1168 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1172 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1176 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1177 dr7 = vcpu->arch.guest_debug_dr7;
1179 dr7 = vcpu->arch.dr7;
1180 static_call(kvm_x86_set_dr7)(vcpu, dr7);
1181 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1182 if (dr7 & DR7_BP_EN_MASK)
1183 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1185 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1187 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1189 u64 fixed = DR6_FIXED_1;
1191 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1194 if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT))
1195 fixed |= DR6_BUS_LOCK;
1199 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1201 size_t size = ARRAY_SIZE(vcpu->arch.db);
1205 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1206 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1207 vcpu->arch.eff_db[dr] = val;
1211 if (!kvm_dr6_valid(val))
1213 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1217 if (!kvm_dr7_valid(val))
1219 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1220 kvm_update_dr7(vcpu);
1226 EXPORT_SYMBOL_GPL(kvm_set_dr);
1228 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1230 size_t size = ARRAY_SIZE(vcpu->arch.db);
1234 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1238 *val = vcpu->arch.dr6;
1242 *val = vcpu->arch.dr7;
1246 EXPORT_SYMBOL_GPL(kvm_get_dr);
1248 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1250 u32 ecx = kvm_rcx_read(vcpu);
1253 if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1254 kvm_inject_gp(vcpu, 0);
1258 kvm_rax_write(vcpu, (u32)data);
1259 kvm_rdx_write(vcpu, data >> 32);
1260 return kvm_skip_emulated_instruction(vcpu);
1262 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1265 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1266 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1268 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1269 * extract the supported MSRs from the related const lists.
1270 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1271 * capabilities of the host cpu. This capabilities test skips MSRs that are
1272 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1273 * may depend on host virtualization features rather than host cpu features.
1276 static const u32 msrs_to_save_all[] = {
1277 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1279 #ifdef CONFIG_X86_64
1280 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1282 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1283 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1285 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1286 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1287 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1288 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1289 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1290 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1291 MSR_IA32_UMWAIT_CONTROL,
1293 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1294 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1295 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1296 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1297 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1298 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1299 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1300 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1301 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1302 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1303 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1304 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1305 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1306 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1307 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1308 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1309 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1310 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1311 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1312 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1313 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1314 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1316 MSR_K7_EVNTSEL0, MSR_K7_EVNTSEL1, MSR_K7_EVNTSEL2, MSR_K7_EVNTSEL3,
1317 MSR_K7_PERFCTR0, MSR_K7_PERFCTR1, MSR_K7_PERFCTR2, MSR_K7_PERFCTR3,
1318 MSR_F15H_PERF_CTL0, MSR_F15H_PERF_CTL1, MSR_F15H_PERF_CTL2,
1319 MSR_F15H_PERF_CTL3, MSR_F15H_PERF_CTL4, MSR_F15H_PERF_CTL5,
1320 MSR_F15H_PERF_CTR0, MSR_F15H_PERF_CTR1, MSR_F15H_PERF_CTR2,
1321 MSR_F15H_PERF_CTR3, MSR_F15H_PERF_CTR4, MSR_F15H_PERF_CTR5,
1324 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1325 static unsigned num_msrs_to_save;
1327 static const u32 emulated_msrs_all[] = {
1328 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1329 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1330 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1331 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1332 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1333 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1334 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1336 HV_X64_MSR_VP_INDEX,
1337 HV_X64_MSR_VP_RUNTIME,
1338 HV_X64_MSR_SCONTROL,
1339 HV_X64_MSR_STIMER0_CONFIG,
1340 HV_X64_MSR_VP_ASSIST_PAGE,
1341 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1342 HV_X64_MSR_TSC_EMULATION_STATUS,
1343 HV_X64_MSR_SYNDBG_OPTIONS,
1344 HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1345 HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1346 HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1348 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1349 MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1351 MSR_IA32_TSC_ADJUST,
1352 MSR_IA32_TSC_DEADLINE,
1353 MSR_IA32_ARCH_CAPABILITIES,
1354 MSR_IA32_PERF_CAPABILITIES,
1355 MSR_IA32_MISC_ENABLE,
1356 MSR_IA32_MCG_STATUS,
1358 MSR_IA32_MCG_EXT_CTL,
1362 MSR_MISC_FEATURES_ENABLES,
1363 MSR_AMD64_VIRT_SPEC_CTRL,
1368 * The following list leaves out MSRs whose values are determined
1369 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1370 * We always support the "true" VMX control MSRs, even if the host
1371 * processor does not, so I am putting these registers here rather
1372 * than in msrs_to_save_all.
1375 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1376 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1377 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1378 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1380 MSR_IA32_VMX_CR0_FIXED0,
1381 MSR_IA32_VMX_CR4_FIXED0,
1382 MSR_IA32_VMX_VMCS_ENUM,
1383 MSR_IA32_VMX_PROCBASED_CTLS2,
1384 MSR_IA32_VMX_EPT_VPID_CAP,
1385 MSR_IA32_VMX_VMFUNC,
1388 MSR_KVM_POLL_CONTROL,
1391 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1392 static unsigned num_emulated_msrs;
1395 * List of msr numbers which are used to expose MSR-based features that
1396 * can be used by a hypervisor to validate requested CPU features.
1398 static const u32 msr_based_features_all[] = {
1400 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1401 MSR_IA32_VMX_PINBASED_CTLS,
1402 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1403 MSR_IA32_VMX_PROCBASED_CTLS,
1404 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1405 MSR_IA32_VMX_EXIT_CTLS,
1406 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1407 MSR_IA32_VMX_ENTRY_CTLS,
1409 MSR_IA32_VMX_CR0_FIXED0,
1410 MSR_IA32_VMX_CR0_FIXED1,
1411 MSR_IA32_VMX_CR4_FIXED0,
1412 MSR_IA32_VMX_CR4_FIXED1,
1413 MSR_IA32_VMX_VMCS_ENUM,
1414 MSR_IA32_VMX_PROCBASED_CTLS2,
1415 MSR_IA32_VMX_EPT_VPID_CAP,
1416 MSR_IA32_VMX_VMFUNC,
1420 MSR_IA32_ARCH_CAPABILITIES,
1421 MSR_IA32_PERF_CAPABILITIES,
1424 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1425 static unsigned int num_msr_based_features;
1427 static u64 kvm_get_arch_capabilities(void)
1431 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1432 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1435 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1436 * the nested hypervisor runs with NX huge pages. If it is not,
1437 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1438 * L1 guests, so it need not worry about its own (L2) guests.
1440 data |= ARCH_CAP_PSCHANGE_MC_NO;
1443 * If we're doing cache flushes (either "always" or "cond")
1444 * we will do one whenever the guest does a vmlaunch/vmresume.
1445 * If an outer hypervisor is doing the cache flush for us
1446 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1447 * capability to the guest too, and if EPT is disabled we're not
1448 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1449 * require a nested hypervisor to do a flush of its own.
1451 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1452 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1454 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1455 data |= ARCH_CAP_RDCL_NO;
1456 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1457 data |= ARCH_CAP_SSB_NO;
1458 if (!boot_cpu_has_bug(X86_BUG_MDS))
1459 data |= ARCH_CAP_MDS_NO;
1461 if (!boot_cpu_has(X86_FEATURE_RTM)) {
1463 * If RTM=0 because the kernel has disabled TSX, the host might
1464 * have TAA_NO or TSX_CTRL. Clear TAA_NO (the guest sees RTM=0
1465 * and therefore knows that there cannot be TAA) but keep
1466 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1467 * and we want to allow migrating those guests to tsx=off hosts.
1469 data &= ~ARCH_CAP_TAA_NO;
1470 } else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1471 data |= ARCH_CAP_TAA_NO;
1474 * Nothing to do here; we emulate TSX_CTRL if present on the
1475 * host so the guest can choose between disabling TSX or
1476 * using VERW to clear CPU buffers.
1483 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1485 switch (msr->index) {
1486 case MSR_IA32_ARCH_CAPABILITIES:
1487 msr->data = kvm_get_arch_capabilities();
1489 case MSR_IA32_UCODE_REV:
1490 rdmsrl_safe(msr->index, &msr->data);
1493 return static_call(kvm_x86_get_msr_feature)(msr);
1498 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1500 struct kvm_msr_entry msr;
1504 r = kvm_get_msr_feature(&msr);
1506 if (r == KVM_MSR_RET_INVALID) {
1507 /* Unconditionally clear the output for simplicity */
1509 if (kvm_msr_ignored_check(index, 0, false))
1521 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1523 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1526 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1529 if (efer & (EFER_LME | EFER_LMA) &&
1530 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1533 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1539 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1541 if (efer & efer_reserved_bits)
1544 return __kvm_valid_efer(vcpu, efer);
1546 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1548 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1550 u64 old_efer = vcpu->arch.efer;
1551 u64 efer = msr_info->data;
1554 if (efer & efer_reserved_bits)
1557 if (!msr_info->host_initiated) {
1558 if (!__kvm_valid_efer(vcpu, efer))
1561 if (is_paging(vcpu) &&
1562 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1567 efer |= vcpu->arch.efer & EFER_LMA;
1569 r = static_call(kvm_x86_set_efer)(vcpu, efer);
1575 /* Update reserved bits */
1576 if ((efer ^ old_efer) & EFER_NX)
1577 kvm_mmu_reset_context(vcpu);
1582 void kvm_enable_efer_bits(u64 mask)
1584 efer_reserved_bits &= ~mask;
1586 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1588 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1590 struct kvm_x86_msr_filter *msr_filter;
1591 struct msr_bitmap_range *ranges;
1592 struct kvm *kvm = vcpu->kvm;
1597 /* x2APIC MSRs do not support filtering. */
1598 if (index >= 0x800 && index <= 0x8ff)
1601 idx = srcu_read_lock(&kvm->srcu);
1603 msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1609 allowed = msr_filter->default_allow;
1610 ranges = msr_filter->ranges;
1612 for (i = 0; i < msr_filter->count; i++) {
1613 u32 start = ranges[i].base;
1614 u32 end = start + ranges[i].nmsrs;
1615 u32 flags = ranges[i].flags;
1616 unsigned long *bitmap = ranges[i].bitmap;
1618 if ((index >= start) && (index < end) && (flags & type)) {
1619 allowed = !!test_bit(index - start, bitmap);
1625 srcu_read_unlock(&kvm->srcu, idx);
1629 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1632 * Write @data into the MSR specified by @index. Select MSR specific fault
1633 * checks are bypassed if @host_initiated is %true.
1634 * Returns 0 on success, non-0 otherwise.
1635 * Assumes vcpu_load() was already called.
1637 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1638 bool host_initiated)
1640 struct msr_data msr;
1642 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1643 return KVM_MSR_RET_FILTERED;
1648 case MSR_KERNEL_GS_BASE:
1651 if (is_noncanonical_address(data, vcpu))
1654 case MSR_IA32_SYSENTER_EIP:
1655 case MSR_IA32_SYSENTER_ESP:
1657 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1658 * non-canonical address is written on Intel but not on
1659 * AMD (which ignores the top 32-bits, because it does
1660 * not implement 64-bit SYSENTER).
1662 * 64-bit code should hence be able to write a non-canonical
1663 * value on AMD. Making the address canonical ensures that
1664 * vmentry does not fail on Intel after writing a non-canonical
1665 * value, and that something deterministic happens if the guest
1666 * invokes 64-bit SYSENTER.
1668 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1671 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1674 if (!host_initiated &&
1675 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1676 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1680 * Per Intel's SDM, bits 63:32 are reserved, but AMD's APM has
1681 * incomplete and conflicting architectural behavior. Current
1682 * AMD CPUs completely ignore bits 63:32, i.e. they aren't
1683 * reserved and always read as zeros. Enforce Intel's reserved
1684 * bits check if and only if the guest CPU is Intel, and clear
1685 * the bits in all other cases. This ensures cross-vendor
1686 * migration will provide consistent behavior for the guest.
1688 if (guest_cpuid_is_intel(vcpu) && (data >> 32) != 0)
1697 msr.host_initiated = host_initiated;
1699 return static_call(kvm_x86_set_msr)(vcpu, &msr);
1702 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1703 u32 index, u64 data, bool host_initiated)
1705 int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1707 if (ret == KVM_MSR_RET_INVALID)
1708 if (kvm_msr_ignored_check(index, data, true))
1715 * Read the MSR specified by @index into @data. Select MSR specific fault
1716 * checks are bypassed if @host_initiated is %true.
1717 * Returns 0 on success, non-0 otherwise.
1718 * Assumes vcpu_load() was already called.
1720 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1721 bool host_initiated)
1723 struct msr_data msr;
1726 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1727 return KVM_MSR_RET_FILTERED;
1731 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1734 if (!host_initiated &&
1735 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1736 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1742 msr.host_initiated = host_initiated;
1744 ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1750 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1751 u32 index, u64 *data, bool host_initiated)
1753 int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1755 if (ret == KVM_MSR_RET_INVALID) {
1756 /* Unconditionally clear *data for simplicity */
1758 if (kvm_msr_ignored_check(index, 0, false))
1765 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1767 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1769 EXPORT_SYMBOL_GPL(kvm_get_msr);
1771 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1773 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1775 EXPORT_SYMBOL_GPL(kvm_set_msr);
1777 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1779 int err = vcpu->run->msr.error;
1781 kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1782 kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1785 return static_call(kvm_x86_complete_emulated_msr)(vcpu, err);
1788 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1790 return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1793 static u64 kvm_msr_reason(int r)
1796 case KVM_MSR_RET_INVALID:
1797 return KVM_MSR_EXIT_REASON_UNKNOWN;
1798 case KVM_MSR_RET_FILTERED:
1799 return KVM_MSR_EXIT_REASON_FILTER;
1801 return KVM_MSR_EXIT_REASON_INVAL;
1805 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1806 u32 exit_reason, u64 data,
1807 int (*completion)(struct kvm_vcpu *vcpu),
1810 u64 msr_reason = kvm_msr_reason(r);
1812 /* Check if the user wanted to know about this MSR fault */
1813 if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1816 vcpu->run->exit_reason = exit_reason;
1817 vcpu->run->msr.error = 0;
1818 memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1819 vcpu->run->msr.reason = msr_reason;
1820 vcpu->run->msr.index = index;
1821 vcpu->run->msr.data = data;
1822 vcpu->arch.complete_userspace_io = completion;
1827 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1829 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1830 complete_emulated_rdmsr, r);
1833 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1835 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1836 complete_emulated_wrmsr, r);
1839 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1841 u32 ecx = kvm_rcx_read(vcpu);
1845 r = kvm_get_msr(vcpu, ecx, &data);
1847 /* MSR read failed? See if we should ask user space */
1848 if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1849 /* Bounce to user space */
1854 trace_kvm_msr_read(ecx, data);
1856 kvm_rax_write(vcpu, data & -1u);
1857 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1859 trace_kvm_msr_read_ex(ecx);
1862 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1864 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1866 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1868 u32 ecx = kvm_rcx_read(vcpu);
1869 u64 data = kvm_read_edx_eax(vcpu);
1872 r = kvm_set_msr(vcpu, ecx, data);
1874 /* MSR write failed? See if we should ask user space */
1875 if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1876 /* Bounce to user space */
1879 /* Signal all other negative errors to userspace */
1884 trace_kvm_msr_write(ecx, data);
1886 trace_kvm_msr_write_ex(ecx, data);
1888 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1890 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1892 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
1894 return kvm_skip_emulated_instruction(vcpu);
1896 EXPORT_SYMBOL_GPL(kvm_emulate_as_nop);
1898 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
1900 /* Treat an INVD instruction as a NOP and just skip it. */
1901 return kvm_emulate_as_nop(vcpu);
1903 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
1905 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
1907 pr_warn_once("kvm: MWAIT instruction emulated as NOP!\n");
1908 return kvm_emulate_as_nop(vcpu);
1910 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
1912 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
1914 kvm_queue_exception(vcpu, UD_VECTOR);
1917 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
1919 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
1921 pr_warn_once("kvm: MONITOR instruction emulated as NOP!\n");
1922 return kvm_emulate_as_nop(vcpu);
1924 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
1926 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1928 xfer_to_guest_mode_prepare();
1929 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1930 xfer_to_guest_mode_work_pending();
1934 * The fast path for frequent and performance sensitive wrmsr emulation,
1935 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1936 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1937 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1938 * other cases which must be called after interrupts are enabled on the host.
1940 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1942 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1945 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1946 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1947 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1948 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
1951 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1952 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1953 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1954 trace_kvm_apic_write(APIC_ICR, (u32)data);
1961 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1963 if (!kvm_can_use_hv_timer(vcpu))
1966 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1970 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1972 u32 msr = kvm_rcx_read(vcpu);
1974 fastpath_t ret = EXIT_FASTPATH_NONE;
1977 case APIC_BASE_MSR + (APIC_ICR >> 4):
1978 data = kvm_read_edx_eax(vcpu);
1979 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1980 kvm_skip_emulated_instruction(vcpu);
1981 ret = EXIT_FASTPATH_EXIT_HANDLED;
1984 case MSR_IA32_TSC_DEADLINE:
1985 data = kvm_read_edx_eax(vcpu);
1986 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1987 kvm_skip_emulated_instruction(vcpu);
1988 ret = EXIT_FASTPATH_REENTER_GUEST;
1995 if (ret != EXIT_FASTPATH_NONE)
1996 trace_kvm_msr_write(msr, data);
2000 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
2003 * Adapt set_msr() to msr_io()'s calling convention
2005 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2007 return kvm_get_msr_ignored_check(vcpu, index, data, true);
2010 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2012 return kvm_set_msr_ignored_check(vcpu, index, *data, true);
2015 #ifdef CONFIG_X86_64
2016 struct pvclock_clock {
2026 struct pvclock_gtod_data {
2029 struct pvclock_clock clock; /* extract of a clocksource struct */
2030 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
2036 static struct pvclock_gtod_data pvclock_gtod_data;
2038 static void update_pvclock_gtod(struct timekeeper *tk)
2040 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
2042 write_seqcount_begin(&vdata->seq);
2044 /* copy pvclock gtod data */
2045 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
2046 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
2047 vdata->clock.mask = tk->tkr_mono.mask;
2048 vdata->clock.mult = tk->tkr_mono.mult;
2049 vdata->clock.shift = tk->tkr_mono.shift;
2050 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
2051 vdata->clock.offset = tk->tkr_mono.base;
2053 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
2054 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
2055 vdata->raw_clock.mask = tk->tkr_raw.mask;
2056 vdata->raw_clock.mult = tk->tkr_raw.mult;
2057 vdata->raw_clock.shift = tk->tkr_raw.shift;
2058 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
2059 vdata->raw_clock.offset = tk->tkr_raw.base;
2061 vdata->wall_time_sec = tk->xtime_sec;
2063 vdata->offs_boot = tk->offs_boot;
2065 write_seqcount_end(&vdata->seq);
2068 static s64 get_kvmclock_base_ns(void)
2070 /* Count up from boot time, but with the frequency of the raw clock. */
2071 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2074 static s64 get_kvmclock_base_ns(void)
2076 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
2077 return ktime_get_boottime_ns();
2081 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2085 struct pvclock_wall_clock wc;
2092 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2097 ++version; /* first time write, random junk */
2101 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2105 * The guest calculates current wall clock time by adding
2106 * system time (updated by kvm_guest_time_update below) to the
2107 * wall clock specified here. We do the reverse here.
2109 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2111 wc.nsec = do_div(wall_nsec, 1000000000);
2112 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2113 wc.version = version;
2115 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2118 wc_sec_hi = wall_nsec >> 32;
2119 kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2120 &wc_sec_hi, sizeof(wc_sec_hi));
2124 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2127 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2128 bool old_msr, bool host_initiated)
2130 struct kvm_arch *ka = &vcpu->kvm->arch;
2132 if (vcpu->vcpu_id == 0 && !host_initiated) {
2133 if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2134 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2136 ka->boot_vcpu_runs_old_kvmclock = old_msr;
2139 vcpu->arch.time = system_time;
2140 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2142 /* we verify if the enable bit is set... */
2143 vcpu->arch.pv_time_enabled = false;
2144 if (!(system_time & 1))
2147 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2148 &vcpu->arch.pv_time, system_time & ~1ULL,
2149 sizeof(struct pvclock_vcpu_time_info)))
2150 vcpu->arch.pv_time_enabled = true;
2155 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2157 do_shl32_div32(dividend, divisor);
2161 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2162 s8 *pshift, u32 *pmultiplier)
2170 scaled64 = scaled_hz;
2171 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2176 tps32 = (uint32_t)tps64;
2177 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2178 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2186 *pmultiplier = div_frac(scaled64, tps32);
2189 #ifdef CONFIG_X86_64
2190 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2193 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2194 static unsigned long max_tsc_khz;
2196 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2198 u64 v = (u64)khz * (1000000 + ppm);
2203 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier);
2205 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2209 /* Guest TSC same frequency as host TSC? */
2211 kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2215 /* TSC scaling supported? */
2216 if (!kvm_has_tsc_control) {
2217 if (user_tsc_khz > tsc_khz) {
2218 vcpu->arch.tsc_catchup = 1;
2219 vcpu->arch.tsc_always_catchup = 1;
2222 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2227 /* TSC scaling required - calculate ratio */
2228 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2229 user_tsc_khz, tsc_khz);
2231 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2232 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2237 kvm_vcpu_write_tsc_multiplier(vcpu, ratio);
2241 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2243 u32 thresh_lo, thresh_hi;
2244 int use_scaling = 0;
2246 /* tsc_khz can be zero if TSC calibration fails */
2247 if (user_tsc_khz == 0) {
2248 /* set tsc_scaling_ratio to a safe value */
2249 kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2253 /* Compute a scale to convert nanoseconds in TSC cycles */
2254 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2255 &vcpu->arch.virtual_tsc_shift,
2256 &vcpu->arch.virtual_tsc_mult);
2257 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2260 * Compute the variation in TSC rate which is acceptable
2261 * within the range of tolerance and decide if the
2262 * rate being applied is within that bounds of the hardware
2263 * rate. If so, no scaling or compensation need be done.
2265 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2266 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2267 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2268 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2271 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2274 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2276 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2277 vcpu->arch.virtual_tsc_mult,
2278 vcpu->arch.virtual_tsc_shift);
2279 tsc += vcpu->arch.this_tsc_write;
2283 static inline int gtod_is_based_on_tsc(int mode)
2285 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2288 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2290 #ifdef CONFIG_X86_64
2292 struct kvm_arch *ka = &vcpu->kvm->arch;
2293 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2295 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2296 atomic_read(&vcpu->kvm->online_vcpus));
2299 * Once the masterclock is enabled, always perform request in
2300 * order to update it.
2302 * In order to enable masterclock, the host clocksource must be TSC
2303 * and the vcpus need to have matched TSCs. When that happens,
2304 * perform request to enable masterclock.
2306 if (ka->use_master_clock ||
2307 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2308 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2310 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2311 atomic_read(&vcpu->kvm->online_vcpus),
2312 ka->use_master_clock, gtod->clock.vclock_mode);
2317 * Multiply tsc by a fixed point number represented by ratio.
2319 * The most significant 64-N bits (mult) of ratio represent the
2320 * integral part of the fixed point number; the remaining N bits
2321 * (frac) represent the fractional part, ie. ratio represents a fixed
2322 * point number (mult + frac * 2^(-N)).
2324 * N equals to kvm_tsc_scaling_ratio_frac_bits.
2326 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2328 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2331 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc, u64 ratio)
2335 if (ratio != kvm_default_tsc_scaling_ratio)
2336 _tsc = __scale_tsc(ratio, tsc);
2340 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2342 static u64 kvm_compute_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2346 tsc = kvm_scale_tsc(vcpu, rdtsc(), vcpu->arch.l1_tsc_scaling_ratio);
2348 return target_tsc - tsc;
2351 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2353 return vcpu->arch.l1_tsc_offset +
2354 kvm_scale_tsc(vcpu, host_tsc, vcpu->arch.l1_tsc_scaling_ratio);
2356 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2358 u64 kvm_calc_nested_tsc_offset(u64 l1_offset, u64 l2_offset, u64 l2_multiplier)
2362 if (l2_multiplier == kvm_default_tsc_scaling_ratio)
2363 nested_offset = l1_offset;
2365 nested_offset = mul_s64_u64_shr((s64) l1_offset, l2_multiplier,
2366 kvm_tsc_scaling_ratio_frac_bits);
2368 nested_offset += l2_offset;
2369 return nested_offset;
2371 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_offset);
2373 u64 kvm_calc_nested_tsc_multiplier(u64 l1_multiplier, u64 l2_multiplier)
2375 if (l2_multiplier != kvm_default_tsc_scaling_ratio)
2376 return mul_u64_u64_shr(l1_multiplier, l2_multiplier,
2377 kvm_tsc_scaling_ratio_frac_bits);
2379 return l1_multiplier;
2381 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_multiplier);
2383 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 l1_offset)
2385 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2386 vcpu->arch.l1_tsc_offset,
2389 vcpu->arch.l1_tsc_offset = l1_offset;
2392 * If we are here because L1 chose not to trap WRMSR to TSC then
2393 * according to the spec this should set L1's TSC (as opposed to
2394 * setting L1's offset for L2).
2396 if (is_guest_mode(vcpu))
2397 vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2399 static_call(kvm_x86_get_l2_tsc_offset)(vcpu),
2400 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2402 vcpu->arch.tsc_offset = l1_offset;
2404 static_call(kvm_x86_write_tsc_offset)(vcpu, vcpu->arch.tsc_offset);
2407 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier)
2409 vcpu->arch.l1_tsc_scaling_ratio = l1_multiplier;
2411 /* Userspace is changing the multiplier while L2 is active */
2412 if (is_guest_mode(vcpu))
2413 vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2415 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2417 vcpu->arch.tsc_scaling_ratio = l1_multiplier;
2419 if (kvm_has_tsc_control)
2420 static_call(kvm_x86_write_tsc_multiplier)(
2421 vcpu, vcpu->arch.tsc_scaling_ratio);
2424 static inline bool kvm_check_tsc_unstable(void)
2426 #ifdef CONFIG_X86_64
2428 * TSC is marked unstable when we're running on Hyper-V,
2429 * 'TSC page' clocksource is good.
2431 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2434 return check_tsc_unstable();
2437 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2439 struct kvm *kvm = vcpu->kvm;
2440 u64 offset, ns, elapsed;
2441 unsigned long flags;
2443 bool already_matched;
2444 bool synchronizing = false;
2446 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2447 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2448 ns = get_kvmclock_base_ns();
2449 elapsed = ns - kvm->arch.last_tsc_nsec;
2451 if (vcpu->arch.virtual_tsc_khz) {
2454 * detection of vcpu initialization -- need to sync
2455 * with other vCPUs. This particularly helps to keep
2456 * kvm_clock stable after CPU hotplug
2458 synchronizing = true;
2460 u64 tsc_exp = kvm->arch.last_tsc_write +
2461 nsec_to_cycles(vcpu, elapsed);
2462 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2464 * Special case: TSC write with a small delta (1 second)
2465 * of virtual cycle time against real time is
2466 * interpreted as an attempt to synchronize the CPU.
2468 synchronizing = data < tsc_exp + tsc_hz &&
2469 data + tsc_hz > tsc_exp;
2474 * For a reliable TSC, we can match TSC offsets, and for an unstable
2475 * TSC, we add elapsed time in this computation. We could let the
2476 * compensation code attempt to catch up if we fall behind, but
2477 * it's better to try to match offsets from the beginning.
2479 if (synchronizing &&
2480 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2481 if (!kvm_check_tsc_unstable()) {
2482 offset = kvm->arch.cur_tsc_offset;
2484 u64 delta = nsec_to_cycles(vcpu, elapsed);
2486 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2489 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2492 * We split periods of matched TSC writes into generations.
2493 * For each generation, we track the original measured
2494 * nanosecond time, offset, and write, so if TSCs are in
2495 * sync, we can match exact offset, and if not, we can match
2496 * exact software computation in compute_guest_tsc()
2498 * These values are tracked in kvm->arch.cur_xxx variables.
2500 kvm->arch.cur_tsc_generation++;
2501 kvm->arch.cur_tsc_nsec = ns;
2502 kvm->arch.cur_tsc_write = data;
2503 kvm->arch.cur_tsc_offset = offset;
2508 * We also track th most recent recorded KHZ, write and time to
2509 * allow the matching interval to be extended at each write.
2511 kvm->arch.last_tsc_nsec = ns;
2512 kvm->arch.last_tsc_write = data;
2513 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2515 vcpu->arch.last_guest_tsc = data;
2517 /* Keep track of which generation this VCPU has synchronized to */
2518 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2519 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2520 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2522 kvm_vcpu_write_tsc_offset(vcpu, offset);
2523 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2525 spin_lock_irqsave(&kvm->arch.pvclock_gtod_sync_lock, flags);
2527 kvm->arch.nr_vcpus_matched_tsc = 0;
2528 } else if (!already_matched) {
2529 kvm->arch.nr_vcpus_matched_tsc++;
2532 kvm_track_tsc_matching(vcpu);
2533 spin_unlock_irqrestore(&kvm->arch.pvclock_gtod_sync_lock, flags);
2536 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2539 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2540 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2543 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2545 if (vcpu->arch.l1_tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2546 WARN_ON(adjustment < 0);
2547 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment,
2548 vcpu->arch.l1_tsc_scaling_ratio);
2549 adjust_tsc_offset_guest(vcpu, adjustment);
2552 #ifdef CONFIG_X86_64
2554 static u64 read_tsc(void)
2556 u64 ret = (u64)rdtsc_ordered();
2557 u64 last = pvclock_gtod_data.clock.cycle_last;
2559 if (likely(ret >= last))
2563 * GCC likes to generate cmov here, but this branch is extremely
2564 * predictable (it's just a function of time and the likely is
2565 * very likely) and there's a data dependence, so force GCC
2566 * to generate a branch instead. I don't barrier() because
2567 * we don't actually need a barrier, and if this function
2568 * ever gets inlined it will generate worse code.
2574 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2580 switch (clock->vclock_mode) {
2581 case VDSO_CLOCKMODE_HVCLOCK:
2582 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2584 if (tsc_pg_val != U64_MAX) {
2585 /* TSC page valid */
2586 *mode = VDSO_CLOCKMODE_HVCLOCK;
2587 v = (tsc_pg_val - clock->cycle_last) &
2590 /* TSC page invalid */
2591 *mode = VDSO_CLOCKMODE_NONE;
2594 case VDSO_CLOCKMODE_TSC:
2595 *mode = VDSO_CLOCKMODE_TSC;
2596 *tsc_timestamp = read_tsc();
2597 v = (*tsc_timestamp - clock->cycle_last) &
2601 *mode = VDSO_CLOCKMODE_NONE;
2604 if (*mode == VDSO_CLOCKMODE_NONE)
2605 *tsc_timestamp = v = 0;
2607 return v * clock->mult;
2610 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2612 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2618 seq = read_seqcount_begin(>od->seq);
2619 ns = gtod->raw_clock.base_cycles;
2620 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2621 ns >>= gtod->raw_clock.shift;
2622 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2623 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2629 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2631 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2637 seq = read_seqcount_begin(>od->seq);
2638 ts->tv_sec = gtod->wall_time_sec;
2639 ns = gtod->clock.base_cycles;
2640 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2641 ns >>= gtod->clock.shift;
2642 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2644 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2650 /* returns true if host is using TSC based clocksource */
2651 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2653 /* checked again under seqlock below */
2654 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2657 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2661 /* returns true if host is using TSC based clocksource */
2662 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2665 /* checked again under seqlock below */
2666 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2669 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2675 * Assuming a stable TSC across physical CPUS, and a stable TSC
2676 * across virtual CPUs, the following condition is possible.
2677 * Each numbered line represents an event visible to both
2678 * CPUs at the next numbered event.
2680 * "timespecX" represents host monotonic time. "tscX" represents
2683 * VCPU0 on CPU0 | VCPU1 on CPU1
2685 * 1. read timespec0,tsc0
2686 * 2. | timespec1 = timespec0 + N
2688 * 3. transition to guest | transition to guest
2689 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2690 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2691 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2693 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2696 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2698 * - 0 < N - M => M < N
2700 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2701 * always the case (the difference between two distinct xtime instances
2702 * might be smaller then the difference between corresponding TSC reads,
2703 * when updating guest vcpus pvclock areas).
2705 * To avoid that problem, do not allow visibility of distinct
2706 * system_timestamp/tsc_timestamp values simultaneously: use a master
2707 * copy of host monotonic time values. Update that master copy
2710 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2714 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2716 #ifdef CONFIG_X86_64
2717 struct kvm_arch *ka = &kvm->arch;
2719 bool host_tsc_clocksource, vcpus_matched;
2721 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2722 atomic_read(&kvm->online_vcpus));
2725 * If the host uses TSC clock, then passthrough TSC as stable
2728 host_tsc_clocksource = kvm_get_time_and_clockread(
2729 &ka->master_kernel_ns,
2730 &ka->master_cycle_now);
2732 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2733 && !ka->backwards_tsc_observed
2734 && !ka->boot_vcpu_runs_old_kvmclock;
2736 if (ka->use_master_clock)
2737 atomic_set(&kvm_guest_has_master_clock, 1);
2739 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2740 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2745 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2747 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2750 static void kvm_gen_update_masterclock(struct kvm *kvm)
2752 #ifdef CONFIG_X86_64
2754 struct kvm_vcpu *vcpu;
2755 struct kvm_arch *ka = &kvm->arch;
2756 unsigned long flags;
2758 kvm_hv_invalidate_tsc_page(kvm);
2760 kvm_make_mclock_inprogress_request(kvm);
2762 /* no guest entries from this point */
2763 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2764 pvclock_update_vm_gtod_copy(kvm);
2765 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2767 kvm_for_each_vcpu(i, vcpu, kvm)
2768 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2770 /* guest entries allowed */
2771 kvm_for_each_vcpu(i, vcpu, kvm)
2772 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2776 u64 get_kvmclock_ns(struct kvm *kvm)
2778 struct kvm_arch *ka = &kvm->arch;
2779 struct pvclock_vcpu_time_info hv_clock;
2780 unsigned long flags;
2783 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2784 if (!ka->use_master_clock) {
2785 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2786 return get_kvmclock_base_ns() + ka->kvmclock_offset;
2789 hv_clock.tsc_timestamp = ka->master_cycle_now;
2790 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2791 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2793 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2796 if (__this_cpu_read(cpu_tsc_khz)) {
2797 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2798 &hv_clock.tsc_shift,
2799 &hv_clock.tsc_to_system_mul);
2800 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2802 ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2809 static void kvm_setup_pvclock_page(struct kvm_vcpu *v,
2810 struct gfn_to_hva_cache *cache,
2811 unsigned int offset)
2813 struct kvm_vcpu_arch *vcpu = &v->arch;
2814 struct pvclock_vcpu_time_info guest_hv_clock;
2816 if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache,
2817 &guest_hv_clock, offset, sizeof(guest_hv_clock))))
2820 /* This VCPU is paused, but it's legal for a guest to read another
2821 * VCPU's kvmclock, so we really have to follow the specification where
2822 * it says that version is odd if data is being modified, and even after
2825 * Version field updates must be kept separate. This is because
2826 * kvm_write_guest_cached might use a "rep movs" instruction, and
2827 * writes within a string instruction are weakly ordered. So there
2828 * are three writes overall.
2830 * As a small optimization, only write the version field in the first
2831 * and third write. The vcpu->pv_time cache is still valid, because the
2832 * version field is the first in the struct.
2834 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2836 if (guest_hv_clock.version & 1)
2837 ++guest_hv_clock.version; /* first time write, random junk */
2839 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2840 kvm_write_guest_offset_cached(v->kvm, cache,
2841 &vcpu->hv_clock, offset,
2842 sizeof(vcpu->hv_clock.version));
2846 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2847 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2849 if (vcpu->pvclock_set_guest_stopped_request) {
2850 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2851 vcpu->pvclock_set_guest_stopped_request = false;
2854 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2856 kvm_write_guest_offset_cached(v->kvm, cache,
2857 &vcpu->hv_clock, offset,
2858 sizeof(vcpu->hv_clock));
2862 vcpu->hv_clock.version++;
2863 kvm_write_guest_offset_cached(v->kvm, cache,
2864 &vcpu->hv_clock, offset,
2865 sizeof(vcpu->hv_clock.version));
2868 static int kvm_guest_time_update(struct kvm_vcpu *v)
2870 unsigned long flags, tgt_tsc_khz;
2871 struct kvm_vcpu_arch *vcpu = &v->arch;
2872 struct kvm_arch *ka = &v->kvm->arch;
2874 u64 tsc_timestamp, host_tsc;
2876 bool use_master_clock;
2882 * If the host uses TSC clock, then passthrough TSC as stable
2885 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2886 use_master_clock = ka->use_master_clock;
2887 if (use_master_clock) {
2888 host_tsc = ka->master_cycle_now;
2889 kernel_ns = ka->master_kernel_ns;
2891 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2893 /* Keep irq disabled to prevent changes to the clock */
2894 local_irq_save(flags);
2895 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2896 if (unlikely(tgt_tsc_khz == 0)) {
2897 local_irq_restore(flags);
2898 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2901 if (!use_master_clock) {
2903 kernel_ns = get_kvmclock_base_ns();
2906 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2909 * We may have to catch up the TSC to match elapsed wall clock
2910 * time for two reasons, even if kvmclock is used.
2911 * 1) CPU could have been running below the maximum TSC rate
2912 * 2) Broken TSC compensation resets the base at each VCPU
2913 * entry to avoid unknown leaps of TSC even when running
2914 * again on the same CPU. This may cause apparent elapsed
2915 * time to disappear, and the guest to stand still or run
2918 if (vcpu->tsc_catchup) {
2919 u64 tsc = compute_guest_tsc(v, kernel_ns);
2920 if (tsc > tsc_timestamp) {
2921 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2922 tsc_timestamp = tsc;
2926 local_irq_restore(flags);
2928 /* With all the info we got, fill in the values */
2930 if (kvm_has_tsc_control)
2931 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz,
2932 v->arch.l1_tsc_scaling_ratio);
2934 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2935 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2936 &vcpu->hv_clock.tsc_shift,
2937 &vcpu->hv_clock.tsc_to_system_mul);
2938 vcpu->hw_tsc_khz = tgt_tsc_khz;
2941 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2942 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2943 vcpu->last_guest_tsc = tsc_timestamp;
2945 /* If the host uses TSC clocksource, then it is stable */
2947 if (use_master_clock)
2948 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2950 vcpu->hv_clock.flags = pvclock_flags;
2952 if (vcpu->pv_time_enabled)
2953 kvm_setup_pvclock_page(v, &vcpu->pv_time, 0);
2954 if (vcpu->xen.vcpu_info_set)
2955 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache,
2956 offsetof(struct compat_vcpu_info, time));
2957 if (vcpu->xen.vcpu_time_info_set)
2958 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0);
2960 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2965 * kvmclock updates which are isolated to a given vcpu, such as
2966 * vcpu->cpu migration, should not allow system_timestamp from
2967 * the rest of the vcpus to remain static. Otherwise ntp frequency
2968 * correction applies to one vcpu's system_timestamp but not
2971 * So in those cases, request a kvmclock update for all vcpus.
2972 * We need to rate-limit these requests though, as they can
2973 * considerably slow guests that have a large number of vcpus.
2974 * The time for a remote vcpu to update its kvmclock is bound
2975 * by the delay we use to rate-limit the updates.
2978 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2980 static void kvmclock_update_fn(struct work_struct *work)
2983 struct delayed_work *dwork = to_delayed_work(work);
2984 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2985 kvmclock_update_work);
2986 struct kvm *kvm = container_of(ka, struct kvm, arch);
2987 struct kvm_vcpu *vcpu;
2989 kvm_for_each_vcpu(i, vcpu, kvm) {
2990 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2991 kvm_vcpu_kick(vcpu);
2995 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2997 struct kvm *kvm = v->kvm;
2999 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
3000 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
3001 KVMCLOCK_UPDATE_DELAY);
3004 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
3006 static void kvmclock_sync_fn(struct work_struct *work)
3008 struct delayed_work *dwork = to_delayed_work(work);
3009 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3010 kvmclock_sync_work);
3011 struct kvm *kvm = container_of(ka, struct kvm, arch);
3013 if (!kvmclock_periodic_sync)
3016 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
3017 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
3018 KVMCLOCK_SYNC_PERIOD);
3022 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
3024 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
3026 /* McStatusWrEn enabled? */
3027 if (guest_cpuid_is_amd_or_hygon(vcpu))
3028 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
3033 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3035 u64 mcg_cap = vcpu->arch.mcg_cap;
3036 unsigned bank_num = mcg_cap & 0xff;
3037 u32 msr = msr_info->index;
3038 u64 data = msr_info->data;
3041 case MSR_IA32_MCG_STATUS:
3042 vcpu->arch.mcg_status = data;
3044 case MSR_IA32_MCG_CTL:
3045 if (!(mcg_cap & MCG_CTL_P) &&
3046 (data || !msr_info->host_initiated))
3048 if (data != 0 && data != ~(u64)0)
3050 vcpu->arch.mcg_ctl = data;
3053 if (msr >= MSR_IA32_MC0_CTL &&
3054 msr < MSR_IA32_MCx_CTL(bank_num)) {
3055 u32 offset = array_index_nospec(
3056 msr - MSR_IA32_MC0_CTL,
3057 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3059 /* only 0 or all 1s can be written to IA32_MCi_CTL
3060 * some Linux kernels though clear bit 10 in bank 4 to
3061 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
3062 * this to avoid an uncatched #GP in the guest
3064 if ((offset & 0x3) == 0 &&
3065 data != 0 && (data | (1 << 10)) != ~(u64)0)
3069 if (!msr_info->host_initiated &&
3070 (offset & 0x3) == 1 && data != 0) {
3071 if (!can_set_mci_status(vcpu))
3075 vcpu->arch.mce_banks[offset] = data;
3083 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
3085 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
3087 return (vcpu->arch.apf.msr_en_val & mask) == mask;
3090 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
3092 gpa_t gpa = data & ~0x3f;
3094 /* Bits 4:5 are reserved, Should be zero */
3098 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
3099 (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
3102 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
3103 (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
3106 if (!lapic_in_kernel(vcpu))
3107 return data ? 1 : 0;
3109 vcpu->arch.apf.msr_en_val = data;
3111 if (!kvm_pv_async_pf_enabled(vcpu)) {
3112 kvm_clear_async_pf_completion_queue(vcpu);
3113 kvm_async_pf_hash_reset(vcpu);
3117 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
3121 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
3122 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
3124 kvm_async_pf_wakeup_all(vcpu);
3129 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
3131 /* Bits 8-63 are reserved */
3135 if (!lapic_in_kernel(vcpu))
3138 vcpu->arch.apf.msr_int_val = data;
3140 vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3145 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3147 vcpu->arch.pv_time_enabled = false;
3148 vcpu->arch.time = 0;
3151 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3153 ++vcpu->stat.tlb_flush;
3154 static_call(kvm_x86_tlb_flush_all)(vcpu);
3157 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3159 ++vcpu->stat.tlb_flush;
3163 * A TLB flush on behalf of the guest is equivalent to
3164 * INVPCID(all), toggling CR4.PGE, etc., which requires
3165 * a forced sync of the shadow page tables. Unload the
3166 * entire MMU here and the subsequent load will sync the
3167 * shadow page tables, and also flush the TLB.
3169 kvm_mmu_unload(vcpu);
3173 static_call(kvm_x86_tlb_flush_guest)(vcpu);
3176 static void record_steal_time(struct kvm_vcpu *vcpu)
3178 struct kvm_host_map map;
3179 struct kvm_steal_time *st;
3181 if (kvm_xen_msr_enabled(vcpu->kvm)) {
3182 kvm_xen_runstate_set_running(vcpu);
3186 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3189 /* -EAGAIN is returned in atomic context so we can just return. */
3190 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
3191 &map, &vcpu->arch.st.cache, false))
3195 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3198 * Doing a TLB flush here, on the guest's behalf, can avoid
3201 if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3202 u8 st_preempted = xchg(&st->preempted, 0);
3204 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3205 st_preempted & KVM_VCPU_FLUSH_TLB);
3206 if (st_preempted & KVM_VCPU_FLUSH_TLB)
3207 kvm_vcpu_flush_tlb_guest(vcpu);
3212 vcpu->arch.st.preempted = 0;
3214 if (st->version & 1)
3215 st->version += 1; /* first time write, random junk */
3221 st->steal += current->sched_info.run_delay -
3222 vcpu->arch.st.last_steal;
3223 vcpu->arch.st.last_steal = current->sched_info.run_delay;
3229 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
3232 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3235 u32 msr = msr_info->index;
3236 u64 data = msr_info->data;
3238 if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3239 return kvm_xen_write_hypercall_page(vcpu, data);
3242 case MSR_AMD64_NB_CFG:
3243 case MSR_IA32_UCODE_WRITE:
3244 case MSR_VM_HSAVE_PA:
3245 case MSR_AMD64_PATCH_LOADER:
3246 case MSR_AMD64_BU_CFG2:
3247 case MSR_AMD64_DC_CFG:
3248 case MSR_F15H_EX_CFG:
3251 case MSR_IA32_UCODE_REV:
3252 if (msr_info->host_initiated)
3253 vcpu->arch.microcode_version = data;
3255 case MSR_IA32_ARCH_CAPABILITIES:
3256 if (!msr_info->host_initiated)
3258 vcpu->arch.arch_capabilities = data;
3260 case MSR_IA32_PERF_CAPABILITIES: {
3261 struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3263 if (!msr_info->host_initiated)
3265 if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3267 if (data & ~msr_ent.data)
3270 vcpu->arch.perf_capabilities = data;
3275 return set_efer(vcpu, msr_info);
3277 data &= ~(u64)0x40; /* ignore flush filter disable */
3278 data &= ~(u64)0x100; /* ignore ignne emulation enable */
3279 data &= ~(u64)0x8; /* ignore TLB cache disable */
3281 /* Handle McStatusWrEn */
3282 if (data == BIT_ULL(18)) {
3283 vcpu->arch.msr_hwcr = data;
3284 } else if (data != 0) {
3285 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3290 case MSR_FAM10H_MMIO_CONF_BASE:
3292 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3297 case 0x200 ... 0x2ff:
3298 return kvm_mtrr_set_msr(vcpu, msr, data);
3299 case MSR_IA32_APICBASE:
3300 return kvm_set_apic_base(vcpu, msr_info);
3301 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3302 return kvm_x2apic_msr_write(vcpu, msr, data);
3303 case MSR_IA32_TSC_DEADLINE:
3304 kvm_set_lapic_tscdeadline_msr(vcpu, data);
3306 case MSR_IA32_TSC_ADJUST:
3307 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3308 if (!msr_info->host_initiated) {
3309 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3310 adjust_tsc_offset_guest(vcpu, adj);
3311 /* Before back to guest, tsc_timestamp must be adjusted
3312 * as well, otherwise guest's percpu pvclock time could jump.
3314 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3316 vcpu->arch.ia32_tsc_adjust_msr = data;
3319 case MSR_IA32_MISC_ENABLE:
3320 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3321 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3322 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3324 vcpu->arch.ia32_misc_enable_msr = data;
3325 kvm_update_cpuid_runtime(vcpu);
3327 vcpu->arch.ia32_misc_enable_msr = data;
3330 case MSR_IA32_SMBASE:
3331 if (!msr_info->host_initiated)
3333 vcpu->arch.smbase = data;
3335 case MSR_IA32_POWER_CTL:
3336 vcpu->arch.msr_ia32_power_ctl = data;
3339 if (msr_info->host_initiated) {
3340 kvm_synchronize_tsc(vcpu, data);
3342 u64 adj = kvm_compute_l1_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3343 adjust_tsc_offset_guest(vcpu, adj);
3344 vcpu->arch.ia32_tsc_adjust_msr += adj;
3348 if (!msr_info->host_initiated &&
3349 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3352 * KVM supports exposing PT to the guest, but does not support
3353 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3354 * XSAVES/XRSTORS to save/restore PT MSRs.
3356 if (data & ~supported_xss)
3358 vcpu->arch.ia32_xss = data;
3361 if (!msr_info->host_initiated)
3363 vcpu->arch.smi_count = data;
3365 case MSR_KVM_WALL_CLOCK_NEW:
3366 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3369 vcpu->kvm->arch.wall_clock = data;
3370 kvm_write_wall_clock(vcpu->kvm, data, 0);
3372 case MSR_KVM_WALL_CLOCK:
3373 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3376 vcpu->kvm->arch.wall_clock = data;
3377 kvm_write_wall_clock(vcpu->kvm, data, 0);
3379 case MSR_KVM_SYSTEM_TIME_NEW:
3380 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3383 kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3385 case MSR_KVM_SYSTEM_TIME:
3386 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3389 kvm_write_system_time(vcpu, data, true, msr_info->host_initiated);
3391 case MSR_KVM_ASYNC_PF_EN:
3392 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3395 if (kvm_pv_enable_async_pf(vcpu, data))
3398 case MSR_KVM_ASYNC_PF_INT:
3399 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3402 if (kvm_pv_enable_async_pf_int(vcpu, data))
3405 case MSR_KVM_ASYNC_PF_ACK:
3406 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3409 vcpu->arch.apf.pageready_pending = false;
3410 kvm_check_async_pf_completion(vcpu);
3413 case MSR_KVM_STEAL_TIME:
3414 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3417 if (unlikely(!sched_info_on()))
3420 if (data & KVM_STEAL_RESERVED_MASK)
3423 vcpu->arch.st.msr_val = data;
3425 if (!(data & KVM_MSR_ENABLED))
3428 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3431 case MSR_KVM_PV_EOI_EN:
3432 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3435 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3439 case MSR_KVM_POLL_CONTROL:
3440 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3443 /* only enable bit supported */
3444 if (data & (-1ULL << 1))
3447 vcpu->arch.msr_kvm_poll_control = data;
3450 case MSR_IA32_MCG_CTL:
3451 case MSR_IA32_MCG_STATUS:
3452 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3453 return set_msr_mce(vcpu, msr_info);
3455 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3456 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3459 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3460 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3461 if (kvm_pmu_is_valid_msr(vcpu, msr))
3462 return kvm_pmu_set_msr(vcpu, msr_info);
3464 if (pr || data != 0)
3465 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3466 "0x%x data 0x%llx\n", msr, data);
3468 case MSR_K7_CLK_CTL:
3470 * Ignore all writes to this no longer documented MSR.
3471 * Writes are only relevant for old K7 processors,
3472 * all pre-dating SVM, but a recommended workaround from
3473 * AMD for these chips. It is possible to specify the
3474 * affected processor models on the command line, hence
3475 * the need to ignore the workaround.
3478 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3479 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3480 case HV_X64_MSR_SYNDBG_OPTIONS:
3481 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3482 case HV_X64_MSR_CRASH_CTL:
3483 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3484 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3485 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3486 case HV_X64_MSR_TSC_EMULATION_STATUS:
3487 return kvm_hv_set_msr_common(vcpu, msr, data,
3488 msr_info->host_initiated);
3489 case MSR_IA32_BBL_CR_CTL3:
3490 /* Drop writes to this legacy MSR -- see rdmsr
3491 * counterpart for further detail.
3493 if (report_ignored_msrs)
3494 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3497 case MSR_AMD64_OSVW_ID_LENGTH:
3498 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3500 vcpu->arch.osvw.length = data;
3502 case MSR_AMD64_OSVW_STATUS:
3503 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3505 vcpu->arch.osvw.status = data;
3507 case MSR_PLATFORM_INFO:
3508 if (!msr_info->host_initiated ||
3509 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3510 cpuid_fault_enabled(vcpu)))
3512 vcpu->arch.msr_platform_info = data;
3514 case MSR_MISC_FEATURES_ENABLES:
3515 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3516 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3517 !supports_cpuid_fault(vcpu)))
3519 vcpu->arch.msr_misc_features_enables = data;
3522 if (kvm_pmu_is_valid_msr(vcpu, msr))
3523 return kvm_pmu_set_msr(vcpu, msr_info);
3524 return KVM_MSR_RET_INVALID;
3528 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3530 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3533 u64 mcg_cap = vcpu->arch.mcg_cap;
3534 unsigned bank_num = mcg_cap & 0xff;
3537 case MSR_IA32_P5_MC_ADDR:
3538 case MSR_IA32_P5_MC_TYPE:
3541 case MSR_IA32_MCG_CAP:
3542 data = vcpu->arch.mcg_cap;
3544 case MSR_IA32_MCG_CTL:
3545 if (!(mcg_cap & MCG_CTL_P) && !host)
3547 data = vcpu->arch.mcg_ctl;
3549 case MSR_IA32_MCG_STATUS:
3550 data = vcpu->arch.mcg_status;
3553 if (msr >= MSR_IA32_MC0_CTL &&
3554 msr < MSR_IA32_MCx_CTL(bank_num)) {
3555 u32 offset = array_index_nospec(
3556 msr - MSR_IA32_MC0_CTL,
3557 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3559 data = vcpu->arch.mce_banks[offset];
3568 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3570 switch (msr_info->index) {
3571 case MSR_IA32_PLATFORM_ID:
3572 case MSR_IA32_EBL_CR_POWERON:
3573 case MSR_IA32_LASTBRANCHFROMIP:
3574 case MSR_IA32_LASTBRANCHTOIP:
3575 case MSR_IA32_LASTINTFROMIP:
3576 case MSR_IA32_LASTINTTOIP:
3577 case MSR_AMD64_SYSCFG:
3578 case MSR_K8_TSEG_ADDR:
3579 case MSR_K8_TSEG_MASK:
3580 case MSR_VM_HSAVE_PA:
3581 case MSR_K8_INT_PENDING_MSG:
3582 case MSR_AMD64_NB_CFG:
3583 case MSR_FAM10H_MMIO_CONF_BASE:
3584 case MSR_AMD64_BU_CFG2:
3585 case MSR_IA32_PERF_CTL:
3586 case MSR_AMD64_DC_CFG:
3587 case MSR_F15H_EX_CFG:
3589 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3590 * limit) MSRs. Just return 0, as we do not want to expose the host
3591 * data here. Do not conditionalize this on CPUID, as KVM does not do
3592 * so for existing CPU-specific MSRs.
3594 case MSR_RAPL_POWER_UNIT:
3595 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3596 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3597 case MSR_PKG_ENERGY_STATUS: /* Total package */
3598 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3601 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3602 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3603 return kvm_pmu_get_msr(vcpu, msr_info);
3604 if (!msr_info->host_initiated)
3608 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3609 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3610 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3611 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3612 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3613 return kvm_pmu_get_msr(vcpu, msr_info);
3616 case MSR_IA32_UCODE_REV:
3617 msr_info->data = vcpu->arch.microcode_version;
3619 case MSR_IA32_ARCH_CAPABILITIES:
3620 if (!msr_info->host_initiated &&
3621 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3623 msr_info->data = vcpu->arch.arch_capabilities;
3625 case MSR_IA32_PERF_CAPABILITIES:
3626 if (!msr_info->host_initiated &&
3627 !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3629 msr_info->data = vcpu->arch.perf_capabilities;
3631 case MSR_IA32_POWER_CTL:
3632 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3634 case MSR_IA32_TSC: {
3636 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3637 * even when not intercepted. AMD manual doesn't explicitly
3638 * state this but appears to behave the same.
3640 * On userspace reads and writes, however, we unconditionally
3641 * return L1's TSC value to ensure backwards-compatible
3642 * behavior for migration.
3646 if (msr_info->host_initiated) {
3647 offset = vcpu->arch.l1_tsc_offset;
3648 ratio = vcpu->arch.l1_tsc_scaling_ratio;
3650 offset = vcpu->arch.tsc_offset;
3651 ratio = vcpu->arch.tsc_scaling_ratio;
3654 msr_info->data = kvm_scale_tsc(vcpu, rdtsc(), ratio) + offset;
3658 case 0x200 ... 0x2ff:
3659 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3660 case 0xcd: /* fsb frequency */
3664 * MSR_EBC_FREQUENCY_ID
3665 * Conservative value valid for even the basic CPU models.
3666 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3667 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3668 * and 266MHz for model 3, or 4. Set Core Clock
3669 * Frequency to System Bus Frequency Ratio to 1 (bits
3670 * 31:24) even though these are only valid for CPU
3671 * models > 2, however guests may end up dividing or
3672 * multiplying by zero otherwise.
3674 case MSR_EBC_FREQUENCY_ID:
3675 msr_info->data = 1 << 24;
3677 case MSR_IA32_APICBASE:
3678 msr_info->data = kvm_get_apic_base(vcpu);
3680 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3681 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3682 case MSR_IA32_TSC_DEADLINE:
3683 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3685 case MSR_IA32_TSC_ADJUST:
3686 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3688 case MSR_IA32_MISC_ENABLE:
3689 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3691 case MSR_IA32_SMBASE:
3692 if (!msr_info->host_initiated)
3694 msr_info->data = vcpu->arch.smbase;
3697 msr_info->data = vcpu->arch.smi_count;
3699 case MSR_IA32_PERF_STATUS:
3700 /* TSC increment by tick */
3701 msr_info->data = 1000ULL;
3702 /* CPU multiplier */
3703 msr_info->data |= (((uint64_t)4ULL) << 40);
3706 msr_info->data = vcpu->arch.efer;
3708 case MSR_KVM_WALL_CLOCK:
3709 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3712 msr_info->data = vcpu->kvm->arch.wall_clock;
3714 case MSR_KVM_WALL_CLOCK_NEW:
3715 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3718 msr_info->data = vcpu->kvm->arch.wall_clock;
3720 case MSR_KVM_SYSTEM_TIME:
3721 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3724 msr_info->data = vcpu->arch.time;
3726 case MSR_KVM_SYSTEM_TIME_NEW:
3727 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3730 msr_info->data = vcpu->arch.time;
3732 case MSR_KVM_ASYNC_PF_EN:
3733 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3736 msr_info->data = vcpu->arch.apf.msr_en_val;
3738 case MSR_KVM_ASYNC_PF_INT:
3739 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3742 msr_info->data = vcpu->arch.apf.msr_int_val;
3744 case MSR_KVM_ASYNC_PF_ACK:
3745 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3750 case MSR_KVM_STEAL_TIME:
3751 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3754 msr_info->data = vcpu->arch.st.msr_val;
3756 case MSR_KVM_PV_EOI_EN:
3757 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3760 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3762 case MSR_KVM_POLL_CONTROL:
3763 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3766 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3768 case MSR_IA32_P5_MC_ADDR:
3769 case MSR_IA32_P5_MC_TYPE:
3770 case MSR_IA32_MCG_CAP:
3771 case MSR_IA32_MCG_CTL:
3772 case MSR_IA32_MCG_STATUS:
3773 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3774 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3775 msr_info->host_initiated);
3777 if (!msr_info->host_initiated &&
3778 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3780 msr_info->data = vcpu->arch.ia32_xss;
3782 case MSR_K7_CLK_CTL:
3784 * Provide expected ramp-up count for K7. All other
3785 * are set to zero, indicating minimum divisors for
3788 * This prevents guest kernels on AMD host with CPU
3789 * type 6, model 8 and higher from exploding due to
3790 * the rdmsr failing.
3792 msr_info->data = 0x20000000;
3794 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3795 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3796 case HV_X64_MSR_SYNDBG_OPTIONS:
3797 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3798 case HV_X64_MSR_CRASH_CTL:
3799 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3800 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3801 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3802 case HV_X64_MSR_TSC_EMULATION_STATUS:
3803 return kvm_hv_get_msr_common(vcpu,
3804 msr_info->index, &msr_info->data,
3805 msr_info->host_initiated);
3806 case MSR_IA32_BBL_CR_CTL3:
3807 /* This legacy MSR exists but isn't fully documented in current
3808 * silicon. It is however accessed by winxp in very narrow
3809 * scenarios where it sets bit #19, itself documented as
3810 * a "reserved" bit. Best effort attempt to source coherent
3811 * read data here should the balance of the register be
3812 * interpreted by the guest:
3814 * L2 cache control register 3: 64GB range, 256KB size,
3815 * enabled, latency 0x1, configured
3817 msr_info->data = 0xbe702111;
3819 case MSR_AMD64_OSVW_ID_LENGTH:
3820 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3822 msr_info->data = vcpu->arch.osvw.length;
3824 case MSR_AMD64_OSVW_STATUS:
3825 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3827 msr_info->data = vcpu->arch.osvw.status;
3829 case MSR_PLATFORM_INFO:
3830 if (!msr_info->host_initiated &&
3831 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3833 msr_info->data = vcpu->arch.msr_platform_info;
3835 case MSR_MISC_FEATURES_ENABLES:
3836 msr_info->data = vcpu->arch.msr_misc_features_enables;
3839 msr_info->data = vcpu->arch.msr_hwcr;
3842 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3843 return kvm_pmu_get_msr(vcpu, msr_info);
3844 return KVM_MSR_RET_INVALID;
3848 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3851 * Read or write a bunch of msrs. All parameters are kernel addresses.
3853 * @return number of msrs set successfully.
3855 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3856 struct kvm_msr_entry *entries,
3857 int (*do_msr)(struct kvm_vcpu *vcpu,
3858 unsigned index, u64 *data))
3862 for (i = 0; i < msrs->nmsrs; ++i)
3863 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3870 * Read or write a bunch of msrs. Parameters are user addresses.
3872 * @return number of msrs set successfully.
3874 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3875 int (*do_msr)(struct kvm_vcpu *vcpu,
3876 unsigned index, u64 *data),
3879 struct kvm_msrs msrs;
3880 struct kvm_msr_entry *entries;
3885 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3889 if (msrs.nmsrs >= MAX_IO_MSRS)
3892 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3893 entries = memdup_user(user_msrs->entries, size);
3894 if (IS_ERR(entries)) {
3895 r = PTR_ERR(entries);
3899 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3904 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3915 static inline bool kvm_can_mwait_in_guest(void)
3917 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3918 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3919 boot_cpu_has(X86_FEATURE_ARAT);
3922 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
3923 struct kvm_cpuid2 __user *cpuid_arg)
3925 struct kvm_cpuid2 cpuid;
3929 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3932 r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3937 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3943 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3948 case KVM_CAP_IRQCHIP:
3950 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3951 case KVM_CAP_SET_TSS_ADDR:
3952 case KVM_CAP_EXT_CPUID:
3953 case KVM_CAP_EXT_EMUL_CPUID:
3954 case KVM_CAP_CLOCKSOURCE:
3956 case KVM_CAP_NOP_IO_DELAY:
3957 case KVM_CAP_MP_STATE:
3958 case KVM_CAP_SYNC_MMU:
3959 case KVM_CAP_USER_NMI:
3960 case KVM_CAP_REINJECT_CONTROL:
3961 case KVM_CAP_IRQ_INJECT_STATUS:
3962 case KVM_CAP_IOEVENTFD:
3963 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3965 case KVM_CAP_PIT_STATE2:
3966 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3967 case KVM_CAP_VCPU_EVENTS:
3968 case KVM_CAP_HYPERV:
3969 case KVM_CAP_HYPERV_VAPIC:
3970 case KVM_CAP_HYPERV_SPIN:
3971 case KVM_CAP_HYPERV_SYNIC:
3972 case KVM_CAP_HYPERV_SYNIC2:
3973 case KVM_CAP_HYPERV_VP_INDEX:
3974 case KVM_CAP_HYPERV_EVENTFD:
3975 case KVM_CAP_HYPERV_TLBFLUSH:
3976 case KVM_CAP_HYPERV_SEND_IPI:
3977 case KVM_CAP_HYPERV_CPUID:
3978 case KVM_CAP_HYPERV_ENFORCE_CPUID:
3979 case KVM_CAP_SYS_HYPERV_CPUID:
3980 case KVM_CAP_PCI_SEGMENT:
3981 case KVM_CAP_DEBUGREGS:
3982 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3984 case KVM_CAP_ASYNC_PF:
3985 case KVM_CAP_ASYNC_PF_INT:
3986 case KVM_CAP_GET_TSC_KHZ:
3987 case KVM_CAP_KVMCLOCK_CTRL:
3988 case KVM_CAP_READONLY_MEM:
3989 case KVM_CAP_HYPERV_TIME:
3990 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3991 case KVM_CAP_TSC_DEADLINE_TIMER:
3992 case KVM_CAP_DISABLE_QUIRKS:
3993 case KVM_CAP_SET_BOOT_CPU_ID:
3994 case KVM_CAP_SPLIT_IRQCHIP:
3995 case KVM_CAP_IMMEDIATE_EXIT:
3996 case KVM_CAP_PMU_EVENT_FILTER:
3997 case KVM_CAP_GET_MSR_FEATURES:
3998 case KVM_CAP_MSR_PLATFORM_INFO:
3999 case KVM_CAP_EXCEPTION_PAYLOAD:
4000 case KVM_CAP_SET_GUEST_DEBUG:
4001 case KVM_CAP_LAST_CPU:
4002 case KVM_CAP_X86_USER_SPACE_MSR:
4003 case KVM_CAP_X86_MSR_FILTER:
4004 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4005 #ifdef CONFIG_X86_SGX_KVM
4006 case KVM_CAP_SGX_ATTRIBUTE:
4008 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
4009 case KVM_CAP_SREGS2:
4010 case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
4013 case KVM_CAP_EXIT_HYPERCALL:
4014 r = KVM_EXIT_HYPERCALL_VALID_MASK;
4016 case KVM_CAP_SET_GUEST_DEBUG2:
4017 return KVM_GUESTDBG_VALID_MASK;
4018 #ifdef CONFIG_KVM_XEN
4019 case KVM_CAP_XEN_HVM:
4020 r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
4021 KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
4022 KVM_XEN_HVM_CONFIG_SHARED_INFO;
4023 if (sched_info_on())
4024 r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
4027 case KVM_CAP_SYNC_REGS:
4028 r = KVM_SYNC_X86_VALID_FIELDS;
4030 case KVM_CAP_ADJUST_CLOCK:
4031 r = KVM_CLOCK_TSC_STABLE;
4033 case KVM_CAP_X86_DISABLE_EXITS:
4034 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
4035 KVM_X86_DISABLE_EXITS_CSTATE;
4036 if(kvm_can_mwait_in_guest())
4037 r |= KVM_X86_DISABLE_EXITS_MWAIT;
4039 case KVM_CAP_X86_SMM:
4040 /* SMBASE is usually relocated above 1M on modern chipsets,
4041 * and SMM handlers might indeed rely on 4G segment limits,
4042 * so do not report SMM to be available if real mode is
4043 * emulated via vm86 mode. Still, do not go to great lengths
4044 * to avoid userspace's usage of the feature, because it is a
4045 * fringe case that is not enabled except via specific settings
4046 * of the module parameters.
4048 r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
4051 r = !static_call(kvm_x86_cpu_has_accelerated_tpr)();
4053 case KVM_CAP_NR_VCPUS:
4054 r = KVM_SOFT_MAX_VCPUS;
4056 case KVM_CAP_MAX_VCPUS:
4059 case KVM_CAP_MAX_VCPU_ID:
4060 r = KVM_MAX_VCPU_IDS;
4062 case KVM_CAP_PV_MMU: /* obsolete */
4066 r = KVM_MAX_MCE_BANKS;
4069 r = boot_cpu_has(X86_FEATURE_XSAVE);
4071 case KVM_CAP_TSC_CONTROL:
4072 r = kvm_has_tsc_control;
4074 case KVM_CAP_X2APIC_API:
4075 r = KVM_X2APIC_API_VALID_FLAGS;
4077 case KVM_CAP_NESTED_STATE:
4078 r = kvm_x86_ops.nested_ops->get_state ?
4079 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
4081 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4082 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
4084 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4085 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
4087 case KVM_CAP_SMALLER_MAXPHYADDR:
4088 r = (int) allow_smaller_maxphyaddr;
4090 case KVM_CAP_STEAL_TIME:
4091 r = sched_info_on();
4093 case KVM_CAP_X86_BUS_LOCK_EXIT:
4094 if (kvm_has_bus_lock_exit)
4095 r = KVM_BUS_LOCK_DETECTION_OFF |
4096 KVM_BUS_LOCK_DETECTION_EXIT;
4107 long kvm_arch_dev_ioctl(struct file *filp,
4108 unsigned int ioctl, unsigned long arg)
4110 void __user *argp = (void __user *)arg;
4114 case KVM_GET_MSR_INDEX_LIST: {
4115 struct kvm_msr_list __user *user_msr_list = argp;
4116 struct kvm_msr_list msr_list;
4120 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4123 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
4124 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4127 if (n < msr_list.nmsrs)
4130 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
4131 num_msrs_to_save * sizeof(u32)))
4133 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
4135 num_emulated_msrs * sizeof(u32)))
4140 case KVM_GET_SUPPORTED_CPUID:
4141 case KVM_GET_EMULATED_CPUID: {
4142 struct kvm_cpuid2 __user *cpuid_arg = argp;
4143 struct kvm_cpuid2 cpuid;
4146 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4149 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
4155 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4160 case KVM_X86_GET_MCE_CAP_SUPPORTED:
4162 if (copy_to_user(argp, &kvm_mce_cap_supported,
4163 sizeof(kvm_mce_cap_supported)))
4167 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
4168 struct kvm_msr_list __user *user_msr_list = argp;
4169 struct kvm_msr_list msr_list;
4173 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4176 msr_list.nmsrs = num_msr_based_features;
4177 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4180 if (n < msr_list.nmsrs)
4183 if (copy_to_user(user_msr_list->indices, &msr_based_features,
4184 num_msr_based_features * sizeof(u32)))
4190 r = msr_io(NULL, argp, do_get_msr_feature, 1);
4192 case KVM_GET_SUPPORTED_HV_CPUID:
4193 r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4203 static void wbinvd_ipi(void *garbage)
4208 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4210 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4213 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4215 /* Address WBINVD may be executed by guest */
4216 if (need_emulate_wbinvd(vcpu)) {
4217 if (static_call(kvm_x86_has_wbinvd_exit)())
4218 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4219 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4220 smp_call_function_single(vcpu->cpu,
4221 wbinvd_ipi, NULL, 1);
4224 static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4226 /* Save host pkru register if supported */
4227 vcpu->arch.host_pkru = read_pkru();
4229 /* Apply any externally detected TSC adjustments (due to suspend) */
4230 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4231 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4232 vcpu->arch.tsc_offset_adjustment = 0;
4233 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4236 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4237 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4238 rdtsc() - vcpu->arch.last_host_tsc;
4240 mark_tsc_unstable("KVM discovered backwards TSC");
4242 if (kvm_check_tsc_unstable()) {
4243 u64 offset = kvm_compute_l1_tsc_offset(vcpu,
4244 vcpu->arch.last_guest_tsc);
4245 kvm_vcpu_write_tsc_offset(vcpu, offset);
4246 vcpu->arch.tsc_catchup = 1;
4249 if (kvm_lapic_hv_timer_in_use(vcpu))
4250 kvm_lapic_restart_hv_timer(vcpu);
4253 * On a host with synchronized TSC, there is no need to update
4254 * kvmclock on vcpu->cpu migration
4256 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4257 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4258 if (vcpu->cpu != cpu)
4259 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4263 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4266 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4268 struct kvm_host_map map;
4269 struct kvm_steal_time *st;
4271 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4274 if (vcpu->arch.st.preempted)
4277 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
4278 &vcpu->arch.st.cache, true))
4282 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
4284 st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4286 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
4289 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4293 if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4294 vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4297 * Take the srcu lock as memslots will be accessed to check the gfn
4298 * cache generation against the memslots generation.
4300 idx = srcu_read_lock(&vcpu->kvm->srcu);
4301 if (kvm_xen_msr_enabled(vcpu->kvm))
4302 kvm_xen_runstate_set_preempted(vcpu);
4304 kvm_steal_time_set_preempted(vcpu);
4305 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4307 static_call(kvm_x86_vcpu_put)(vcpu);
4308 vcpu->arch.last_host_tsc = rdtsc();
4311 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4312 struct kvm_lapic_state *s)
4314 if (vcpu->arch.apicv_active)
4315 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
4317 return kvm_apic_get_state(vcpu, s);
4320 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4321 struct kvm_lapic_state *s)
4325 r = kvm_apic_set_state(vcpu, s);
4328 update_cr8_intercept(vcpu);
4333 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4336 * We can accept userspace's request for interrupt injection
4337 * as long as we have a place to store the interrupt number.
4338 * The actual injection will happen when the CPU is able to
4339 * deliver the interrupt.
4341 if (kvm_cpu_has_extint(vcpu))
4344 /* Acknowledging ExtINT does not happen if LINT0 is masked. */
4345 return (!lapic_in_kernel(vcpu) ||
4346 kvm_apic_accept_pic_intr(vcpu));
4349 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4352 * Do not cause an interrupt window exit if an exception
4353 * is pending or an event needs reinjection; userspace
4354 * might want to inject the interrupt manually using KVM_SET_REGS
4355 * or KVM_SET_SREGS. For that to work, we must be at an
4356 * instruction boundary and with no events half-injected.
4358 return (kvm_arch_interrupt_allowed(vcpu) &&
4359 kvm_cpu_accept_dm_intr(vcpu) &&
4360 !kvm_event_needs_reinjection(vcpu) &&
4361 !vcpu->arch.exception.pending);
4364 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4365 struct kvm_interrupt *irq)
4367 if (irq->irq >= KVM_NR_INTERRUPTS)
4370 if (!irqchip_in_kernel(vcpu->kvm)) {
4371 kvm_queue_interrupt(vcpu, irq->irq, false);
4372 kvm_make_request(KVM_REQ_EVENT, vcpu);
4377 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4378 * fail for in-kernel 8259.
4380 if (pic_in_kernel(vcpu->kvm))
4383 if (vcpu->arch.pending_external_vector != -1)
4386 vcpu->arch.pending_external_vector = irq->irq;
4387 kvm_make_request(KVM_REQ_EVENT, vcpu);
4391 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4393 kvm_inject_nmi(vcpu);
4398 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4400 kvm_make_request(KVM_REQ_SMI, vcpu);
4405 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4406 struct kvm_tpr_access_ctl *tac)
4410 vcpu->arch.tpr_access_reporting = !!tac->enabled;
4414 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4418 unsigned bank_num = mcg_cap & 0xff, bank;
4421 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4423 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4426 vcpu->arch.mcg_cap = mcg_cap;
4427 /* Init IA32_MCG_CTL to all 1s */
4428 if (mcg_cap & MCG_CTL_P)
4429 vcpu->arch.mcg_ctl = ~(u64)0;
4430 /* Init IA32_MCi_CTL to all 1s */
4431 for (bank = 0; bank < bank_num; bank++)
4432 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4434 static_call(kvm_x86_setup_mce)(vcpu);
4439 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4440 struct kvm_x86_mce *mce)
4442 u64 mcg_cap = vcpu->arch.mcg_cap;
4443 unsigned bank_num = mcg_cap & 0xff;
4444 u64 *banks = vcpu->arch.mce_banks;
4446 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4449 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4450 * reporting is disabled
4452 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4453 vcpu->arch.mcg_ctl != ~(u64)0)
4455 banks += 4 * mce->bank;
4457 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4458 * reporting is disabled for the bank
4460 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4462 if (mce->status & MCI_STATUS_UC) {
4463 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4464 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4465 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4468 if (banks[1] & MCI_STATUS_VAL)
4469 mce->status |= MCI_STATUS_OVER;
4470 banks[2] = mce->addr;
4471 banks[3] = mce->misc;
4472 vcpu->arch.mcg_status = mce->mcg_status;
4473 banks[1] = mce->status;
4474 kvm_queue_exception(vcpu, MC_VECTOR);
4475 } else if (!(banks[1] & MCI_STATUS_VAL)
4476 || !(banks[1] & MCI_STATUS_UC)) {
4477 if (banks[1] & MCI_STATUS_VAL)
4478 mce->status |= MCI_STATUS_OVER;
4479 banks[2] = mce->addr;
4480 banks[3] = mce->misc;
4481 banks[1] = mce->status;
4483 banks[1] |= MCI_STATUS_OVER;
4487 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4488 struct kvm_vcpu_events *events)
4492 if (kvm_check_request(KVM_REQ_SMI, vcpu))
4496 * In guest mode, payload delivery should be deferred,
4497 * so that the L1 hypervisor can intercept #PF before
4498 * CR2 is modified (or intercept #DB before DR6 is
4499 * modified under nVMX). Unless the per-VM capability,
4500 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4501 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4502 * opportunistically defer the exception payload, deliver it if the
4503 * capability hasn't been requested before processing a
4504 * KVM_GET_VCPU_EVENTS.
4506 if (!vcpu->kvm->arch.exception_payload_enabled &&
4507 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4508 kvm_deliver_exception_payload(vcpu);
4511 * The API doesn't provide the instruction length for software
4512 * exceptions, so don't report them. As long as the guest RIP
4513 * isn't advanced, we should expect to encounter the exception
4516 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4517 events->exception.injected = 0;
4518 events->exception.pending = 0;
4520 events->exception.injected = vcpu->arch.exception.injected;
4521 events->exception.pending = vcpu->arch.exception.pending;
4523 * For ABI compatibility, deliberately conflate
4524 * pending and injected exceptions when
4525 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4527 if (!vcpu->kvm->arch.exception_payload_enabled)
4528 events->exception.injected |=
4529 vcpu->arch.exception.pending;
4531 events->exception.nr = vcpu->arch.exception.nr;
4532 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4533 events->exception.error_code = vcpu->arch.exception.error_code;
4534 events->exception_has_payload = vcpu->arch.exception.has_payload;
4535 events->exception_payload = vcpu->arch.exception.payload;
4537 events->interrupt.injected =
4538 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4539 events->interrupt.nr = vcpu->arch.interrupt.nr;
4540 events->interrupt.soft = 0;
4541 events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4543 events->nmi.injected = vcpu->arch.nmi_injected;
4544 events->nmi.pending = vcpu->arch.nmi_pending != 0;
4545 events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4546 events->nmi.pad = 0;
4548 events->sipi_vector = 0; /* never valid when reporting to user space */
4550 events->smi.smm = is_smm(vcpu);
4551 events->smi.pending = vcpu->arch.smi_pending;
4552 events->smi.smm_inside_nmi =
4553 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4554 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4556 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4557 | KVM_VCPUEVENT_VALID_SHADOW
4558 | KVM_VCPUEVENT_VALID_SMM);
4559 if (vcpu->kvm->arch.exception_payload_enabled)
4560 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4562 memset(&events->reserved, 0, sizeof(events->reserved));
4565 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm);
4567 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4568 struct kvm_vcpu_events *events)
4570 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4571 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4572 | KVM_VCPUEVENT_VALID_SHADOW
4573 | KVM_VCPUEVENT_VALID_SMM
4574 | KVM_VCPUEVENT_VALID_PAYLOAD))
4577 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4578 if (!vcpu->kvm->arch.exception_payload_enabled)
4580 if (events->exception.pending)
4581 events->exception.injected = 0;
4583 events->exception_has_payload = 0;
4585 events->exception.pending = 0;
4586 events->exception_has_payload = 0;
4589 if ((events->exception.injected || events->exception.pending) &&
4590 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4593 /* INITs are latched while in SMM */
4594 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4595 (events->smi.smm || events->smi.pending) &&
4596 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4600 vcpu->arch.exception.injected = events->exception.injected;
4601 vcpu->arch.exception.pending = events->exception.pending;
4602 vcpu->arch.exception.nr = events->exception.nr;
4603 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4604 vcpu->arch.exception.error_code = events->exception.error_code;
4605 vcpu->arch.exception.has_payload = events->exception_has_payload;
4606 vcpu->arch.exception.payload = events->exception_payload;
4608 vcpu->arch.interrupt.injected = events->interrupt.injected;
4609 vcpu->arch.interrupt.nr = events->interrupt.nr;
4610 vcpu->arch.interrupt.soft = events->interrupt.soft;
4611 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4612 static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4613 events->interrupt.shadow);
4615 vcpu->arch.nmi_injected = events->nmi.injected;
4616 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4617 vcpu->arch.nmi_pending = events->nmi.pending;
4618 static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4620 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4621 lapic_in_kernel(vcpu))
4622 vcpu->arch.apic->sipi_vector = events->sipi_vector;
4624 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4625 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm)
4626 kvm_smm_changed(vcpu, events->smi.smm);
4628 vcpu->arch.smi_pending = events->smi.pending;
4630 if (events->smi.smm) {
4631 if (events->smi.smm_inside_nmi)
4632 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4634 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4637 if (lapic_in_kernel(vcpu)) {
4638 if (events->smi.latched_init)
4639 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4641 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4645 kvm_make_request(KVM_REQ_EVENT, vcpu);
4650 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4651 struct kvm_debugregs *dbgregs)
4655 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4656 kvm_get_dr(vcpu, 6, &val);
4658 dbgregs->dr7 = vcpu->arch.dr7;
4660 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4663 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4664 struct kvm_debugregs *dbgregs)
4669 if (!kvm_dr6_valid(dbgregs->dr6))
4671 if (!kvm_dr7_valid(dbgregs->dr7))
4674 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4675 kvm_update_dr0123(vcpu);
4676 vcpu->arch.dr6 = dbgregs->dr6;
4677 vcpu->arch.dr7 = dbgregs->dr7;
4678 kvm_update_dr7(vcpu);
4683 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4685 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4687 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4688 u64 xstate_bv = xsave->header.xfeatures;
4692 * Copy legacy XSAVE area, to avoid complications with CPUID
4693 * leaves 0 and 1 in the loop below.
4695 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4698 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4699 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4702 * Copy each region from the possibly compacted offset to the
4703 * non-compacted offset.
4705 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4707 u32 size, offset, ecx, edx;
4708 u64 xfeature_mask = valid & -valid;
4709 int xfeature_nr = fls64(xfeature_mask) - 1;
4712 cpuid_count(XSTATE_CPUID, xfeature_nr,
4713 &size, &offset, &ecx, &edx);
4715 if (xfeature_nr == XFEATURE_PKRU) {
4716 memcpy(dest + offset, &vcpu->arch.pkru,
4717 sizeof(vcpu->arch.pkru));
4719 src = get_xsave_addr(xsave, xfeature_nr);
4721 memcpy(dest + offset, src, size);
4724 valid -= xfeature_mask;
4728 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4730 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4731 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4735 * Copy legacy XSAVE area, to avoid complications with CPUID
4736 * leaves 0 and 1 in the loop below.
4738 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4740 /* Set XSTATE_BV and possibly XCOMP_BV. */
4741 xsave->header.xfeatures = xstate_bv;
4742 if (boot_cpu_has(X86_FEATURE_XSAVES))
4743 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4746 * Copy each region from the non-compacted offset to the
4747 * possibly compacted offset.
4749 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4751 u32 size, offset, ecx, edx;
4752 u64 xfeature_mask = valid & -valid;
4753 int xfeature_nr = fls64(xfeature_mask) - 1;
4755 cpuid_count(XSTATE_CPUID, xfeature_nr,
4756 &size, &offset, &ecx, &edx);
4758 if (xfeature_nr == XFEATURE_PKRU) {
4759 memcpy(&vcpu->arch.pkru, src + offset,
4760 sizeof(vcpu->arch.pkru));
4762 void *dest = get_xsave_addr(xsave, xfeature_nr);
4765 memcpy(dest, src + offset, size);
4768 valid -= xfeature_mask;
4772 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4773 struct kvm_xsave *guest_xsave)
4775 if (!vcpu->arch.guest_fpu)
4778 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4779 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4780 fill_xsave((u8 *) guest_xsave->region, vcpu);
4782 memcpy(guest_xsave->region,
4783 &vcpu->arch.guest_fpu->state.fxsave,
4784 sizeof(struct fxregs_state));
4785 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4786 XFEATURE_MASK_FPSSE;
4790 #define XSAVE_MXCSR_OFFSET 24
4792 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4793 struct kvm_xsave *guest_xsave)
4798 if (!vcpu->arch.guest_fpu)
4801 xstate_bv = *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4802 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4804 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4806 * Here we allow setting states that are not present in
4807 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4808 * with old userspace.
4810 if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4812 load_xsave(vcpu, (u8 *)guest_xsave->region);
4814 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4815 mxcsr & ~mxcsr_feature_mask)
4817 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4818 guest_xsave->region, sizeof(struct fxregs_state));
4823 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4824 struct kvm_xcrs *guest_xcrs)
4826 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4827 guest_xcrs->nr_xcrs = 0;
4831 guest_xcrs->nr_xcrs = 1;
4832 guest_xcrs->flags = 0;
4833 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4834 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4837 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4838 struct kvm_xcrs *guest_xcrs)
4842 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4845 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4848 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4849 /* Only support XCR0 currently */
4850 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4851 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4852 guest_xcrs->xcrs[i].value);
4861 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4862 * stopped by the hypervisor. This function will be called from the host only.
4863 * EINVAL is returned when the host attempts to set the flag for a guest that
4864 * does not support pv clocks.
4866 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4868 if (!vcpu->arch.pv_time_enabled)
4870 vcpu->arch.pvclock_set_guest_stopped_request = true;
4871 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4875 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4876 struct kvm_enable_cap *cap)
4879 uint16_t vmcs_version;
4880 void __user *user_ptr;
4886 case KVM_CAP_HYPERV_SYNIC2:
4891 case KVM_CAP_HYPERV_SYNIC:
4892 if (!irqchip_in_kernel(vcpu->kvm))
4894 return kvm_hv_activate_synic(vcpu, cap->cap ==
4895 KVM_CAP_HYPERV_SYNIC2);
4896 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4897 if (!kvm_x86_ops.nested_ops->enable_evmcs)
4899 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4901 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4902 if (copy_to_user(user_ptr, &vmcs_version,
4903 sizeof(vmcs_version)))
4907 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4908 if (!kvm_x86_ops.enable_direct_tlbflush)
4911 return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
4913 case KVM_CAP_HYPERV_ENFORCE_CPUID:
4914 return kvm_hv_set_enforce_cpuid(vcpu, cap->args[0]);
4916 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4917 vcpu->arch.pv_cpuid.enforce = cap->args[0];
4918 if (vcpu->arch.pv_cpuid.enforce)
4919 kvm_update_pv_runtime(vcpu);
4927 long kvm_arch_vcpu_ioctl(struct file *filp,
4928 unsigned int ioctl, unsigned long arg)
4930 struct kvm_vcpu *vcpu = filp->private_data;
4931 void __user *argp = (void __user *)arg;
4934 struct kvm_sregs2 *sregs2;
4935 struct kvm_lapic_state *lapic;
4936 struct kvm_xsave *xsave;
4937 struct kvm_xcrs *xcrs;
4945 case KVM_GET_LAPIC: {
4947 if (!lapic_in_kernel(vcpu))
4949 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4950 GFP_KERNEL_ACCOUNT);
4955 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4959 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4964 case KVM_SET_LAPIC: {
4966 if (!lapic_in_kernel(vcpu))
4968 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4969 if (IS_ERR(u.lapic)) {
4970 r = PTR_ERR(u.lapic);
4974 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4977 case KVM_INTERRUPT: {
4978 struct kvm_interrupt irq;
4981 if (copy_from_user(&irq, argp, sizeof(irq)))
4983 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4987 r = kvm_vcpu_ioctl_nmi(vcpu);
4991 r = kvm_vcpu_ioctl_smi(vcpu);
4994 case KVM_SET_CPUID: {
4995 struct kvm_cpuid __user *cpuid_arg = argp;
4996 struct kvm_cpuid cpuid;
4999 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5001 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
5004 case KVM_SET_CPUID2: {
5005 struct kvm_cpuid2 __user *cpuid_arg = argp;
5006 struct kvm_cpuid2 cpuid;
5009 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5011 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
5012 cpuid_arg->entries);
5015 case KVM_GET_CPUID2: {
5016 struct kvm_cpuid2 __user *cpuid_arg = argp;
5017 struct kvm_cpuid2 cpuid;
5020 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5022 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
5023 cpuid_arg->entries);
5027 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
5032 case KVM_GET_MSRS: {
5033 int idx = srcu_read_lock(&vcpu->kvm->srcu);
5034 r = msr_io(vcpu, argp, do_get_msr, 1);
5035 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5038 case KVM_SET_MSRS: {
5039 int idx = srcu_read_lock(&vcpu->kvm->srcu);
5040 r = msr_io(vcpu, argp, do_set_msr, 0);
5041 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5044 case KVM_TPR_ACCESS_REPORTING: {
5045 struct kvm_tpr_access_ctl tac;
5048 if (copy_from_user(&tac, argp, sizeof(tac)))
5050 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
5054 if (copy_to_user(argp, &tac, sizeof(tac)))
5059 case KVM_SET_VAPIC_ADDR: {
5060 struct kvm_vapic_addr va;
5064 if (!lapic_in_kernel(vcpu))
5067 if (copy_from_user(&va, argp, sizeof(va)))
5069 idx = srcu_read_lock(&vcpu->kvm->srcu);
5070 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
5071 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5074 case KVM_X86_SETUP_MCE: {
5078 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
5080 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
5083 case KVM_X86_SET_MCE: {
5084 struct kvm_x86_mce mce;
5087 if (copy_from_user(&mce, argp, sizeof(mce)))
5089 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
5092 case KVM_GET_VCPU_EVENTS: {
5093 struct kvm_vcpu_events events;
5095 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
5098 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
5103 case KVM_SET_VCPU_EVENTS: {
5104 struct kvm_vcpu_events events;
5107 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
5110 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
5113 case KVM_GET_DEBUGREGS: {
5114 struct kvm_debugregs dbgregs;
5116 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
5119 if (copy_to_user(argp, &dbgregs,
5120 sizeof(struct kvm_debugregs)))
5125 case KVM_SET_DEBUGREGS: {
5126 struct kvm_debugregs dbgregs;
5129 if (copy_from_user(&dbgregs, argp,
5130 sizeof(struct kvm_debugregs)))
5133 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
5136 case KVM_GET_XSAVE: {
5137 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
5142 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
5145 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
5150 case KVM_SET_XSAVE: {
5151 u.xsave = memdup_user(argp, sizeof(*u.xsave));
5152 if (IS_ERR(u.xsave)) {
5153 r = PTR_ERR(u.xsave);
5157 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
5160 case KVM_GET_XCRS: {
5161 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5166 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5169 if (copy_to_user(argp, u.xcrs,
5170 sizeof(struct kvm_xcrs)))
5175 case KVM_SET_XCRS: {
5176 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5177 if (IS_ERR(u.xcrs)) {
5178 r = PTR_ERR(u.xcrs);
5182 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5185 case KVM_SET_TSC_KHZ: {
5189 user_tsc_khz = (u32)arg;
5191 if (kvm_has_tsc_control &&
5192 user_tsc_khz >= kvm_max_guest_tsc_khz)
5195 if (user_tsc_khz == 0)
5196 user_tsc_khz = tsc_khz;
5198 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5203 case KVM_GET_TSC_KHZ: {
5204 r = vcpu->arch.virtual_tsc_khz;
5207 case KVM_KVMCLOCK_CTRL: {
5208 r = kvm_set_guest_paused(vcpu);
5211 case KVM_ENABLE_CAP: {
5212 struct kvm_enable_cap cap;
5215 if (copy_from_user(&cap, argp, sizeof(cap)))
5217 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5220 case KVM_GET_NESTED_STATE: {
5221 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5225 if (!kvm_x86_ops.nested_ops->get_state)
5228 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5230 if (get_user(user_data_size, &user_kvm_nested_state->size))
5233 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5238 if (r > user_data_size) {
5239 if (put_user(r, &user_kvm_nested_state->size))
5249 case KVM_SET_NESTED_STATE: {
5250 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5251 struct kvm_nested_state kvm_state;
5255 if (!kvm_x86_ops.nested_ops->set_state)
5259 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5263 if (kvm_state.size < sizeof(kvm_state))
5266 if (kvm_state.flags &
5267 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5268 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5269 | KVM_STATE_NESTED_GIF_SET))
5272 /* nested_run_pending implies guest_mode. */
5273 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5274 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5277 idx = srcu_read_lock(&vcpu->kvm->srcu);
5278 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5279 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5282 case KVM_GET_SUPPORTED_HV_CPUID:
5283 r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5285 #ifdef CONFIG_KVM_XEN
5286 case KVM_XEN_VCPU_GET_ATTR: {
5287 struct kvm_xen_vcpu_attr xva;
5290 if (copy_from_user(&xva, argp, sizeof(xva)))
5292 r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5293 if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5297 case KVM_XEN_VCPU_SET_ATTR: {
5298 struct kvm_xen_vcpu_attr xva;
5301 if (copy_from_user(&xva, argp, sizeof(xva)))
5303 r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5307 case KVM_GET_SREGS2: {
5308 u.sregs2 = kzalloc(sizeof(struct kvm_sregs2), GFP_KERNEL);
5312 __get_sregs2(vcpu, u.sregs2);
5314 if (copy_to_user(argp, u.sregs2, sizeof(struct kvm_sregs2)))
5319 case KVM_SET_SREGS2: {
5320 u.sregs2 = memdup_user(argp, sizeof(struct kvm_sregs2));
5321 if (IS_ERR(u.sregs2)) {
5322 r = PTR_ERR(u.sregs2);
5326 r = __set_sregs2(vcpu, u.sregs2);
5339 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5341 return VM_FAULT_SIGBUS;
5344 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5348 if (addr > (unsigned int)(-3 * PAGE_SIZE))
5350 ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5354 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5357 return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5360 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5361 unsigned long kvm_nr_mmu_pages)
5363 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5366 mutex_lock(&kvm->slots_lock);
5368 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5369 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5371 mutex_unlock(&kvm->slots_lock);
5375 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5377 return kvm->arch.n_max_mmu_pages;
5380 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5382 struct kvm_pic *pic = kvm->arch.vpic;
5386 switch (chip->chip_id) {
5387 case KVM_IRQCHIP_PIC_MASTER:
5388 memcpy(&chip->chip.pic, &pic->pics[0],
5389 sizeof(struct kvm_pic_state));
5391 case KVM_IRQCHIP_PIC_SLAVE:
5392 memcpy(&chip->chip.pic, &pic->pics[1],
5393 sizeof(struct kvm_pic_state));
5395 case KVM_IRQCHIP_IOAPIC:
5396 kvm_get_ioapic(kvm, &chip->chip.ioapic);
5405 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5407 struct kvm_pic *pic = kvm->arch.vpic;
5411 switch (chip->chip_id) {
5412 case KVM_IRQCHIP_PIC_MASTER:
5413 spin_lock(&pic->lock);
5414 memcpy(&pic->pics[0], &chip->chip.pic,
5415 sizeof(struct kvm_pic_state));
5416 spin_unlock(&pic->lock);
5418 case KVM_IRQCHIP_PIC_SLAVE:
5419 spin_lock(&pic->lock);
5420 memcpy(&pic->pics[1], &chip->chip.pic,
5421 sizeof(struct kvm_pic_state));
5422 spin_unlock(&pic->lock);
5424 case KVM_IRQCHIP_IOAPIC:
5425 kvm_set_ioapic(kvm, &chip->chip.ioapic);
5431 kvm_pic_update_irq(pic);
5435 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5437 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5439 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5441 mutex_lock(&kps->lock);
5442 memcpy(ps, &kps->channels, sizeof(*ps));
5443 mutex_unlock(&kps->lock);
5447 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5450 struct kvm_pit *pit = kvm->arch.vpit;
5452 mutex_lock(&pit->pit_state.lock);
5453 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5454 for (i = 0; i < 3; i++)
5455 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5456 mutex_unlock(&pit->pit_state.lock);
5460 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5462 mutex_lock(&kvm->arch.vpit->pit_state.lock);
5463 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5464 sizeof(ps->channels));
5465 ps->flags = kvm->arch.vpit->pit_state.flags;
5466 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5467 memset(&ps->reserved, 0, sizeof(ps->reserved));
5471 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5475 u32 prev_legacy, cur_legacy;
5476 struct kvm_pit *pit = kvm->arch.vpit;
5478 mutex_lock(&pit->pit_state.lock);
5479 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5480 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5481 if (!prev_legacy && cur_legacy)
5483 memcpy(&pit->pit_state.channels, &ps->channels,
5484 sizeof(pit->pit_state.channels));
5485 pit->pit_state.flags = ps->flags;
5486 for (i = 0; i < 3; i++)
5487 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5489 mutex_unlock(&pit->pit_state.lock);
5493 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5494 struct kvm_reinject_control *control)
5496 struct kvm_pit *pit = kvm->arch.vpit;
5498 /* pit->pit_state.lock was overloaded to prevent userspace from getting
5499 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5500 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
5502 mutex_lock(&pit->pit_state.lock);
5503 kvm_pit_set_reinject(pit, control->pit_reinject);
5504 mutex_unlock(&pit->pit_state.lock);
5509 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5513 * Flush all CPUs' dirty log buffers to the dirty_bitmap. Called
5514 * before reporting dirty_bitmap to userspace. KVM flushes the buffers
5515 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5518 struct kvm_vcpu *vcpu;
5521 kvm_for_each_vcpu(i, vcpu, kvm)
5522 kvm_vcpu_kick(vcpu);
5525 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5528 if (!irqchip_in_kernel(kvm))
5531 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5532 irq_event->irq, irq_event->level,
5537 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5538 struct kvm_enable_cap *cap)
5546 case KVM_CAP_DISABLE_QUIRKS:
5547 kvm->arch.disabled_quirks = cap->args[0];
5550 case KVM_CAP_SPLIT_IRQCHIP: {
5551 mutex_lock(&kvm->lock);
5553 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5554 goto split_irqchip_unlock;
5556 if (irqchip_in_kernel(kvm))
5557 goto split_irqchip_unlock;
5558 if (kvm->created_vcpus)
5559 goto split_irqchip_unlock;
5560 r = kvm_setup_empty_irq_routing(kvm);
5562 goto split_irqchip_unlock;
5563 /* Pairs with irqchip_in_kernel. */
5565 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5566 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5568 split_irqchip_unlock:
5569 mutex_unlock(&kvm->lock);
5572 case KVM_CAP_X2APIC_API:
5574 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5577 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5578 kvm->arch.x2apic_format = true;
5579 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5580 kvm->arch.x2apic_broadcast_quirk_disabled = true;
5584 case KVM_CAP_X86_DISABLE_EXITS:
5586 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5589 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5590 kvm_can_mwait_in_guest())
5591 kvm->arch.mwait_in_guest = true;
5592 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5593 kvm->arch.hlt_in_guest = true;
5594 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5595 kvm->arch.pause_in_guest = true;
5596 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5597 kvm->arch.cstate_in_guest = true;
5600 case KVM_CAP_MSR_PLATFORM_INFO:
5601 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5604 case KVM_CAP_EXCEPTION_PAYLOAD:
5605 kvm->arch.exception_payload_enabled = cap->args[0];
5608 case KVM_CAP_X86_USER_SPACE_MSR:
5609 kvm->arch.user_space_msr_mask = cap->args[0];
5612 case KVM_CAP_X86_BUS_LOCK_EXIT:
5614 if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5617 if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5618 (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5621 if (kvm_has_bus_lock_exit &&
5622 cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5623 kvm->arch.bus_lock_detection_enabled = true;
5626 #ifdef CONFIG_X86_SGX_KVM
5627 case KVM_CAP_SGX_ATTRIBUTE: {
5628 unsigned long allowed_attributes = 0;
5630 r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
5634 /* KVM only supports the PROVISIONKEY privileged attribute. */
5635 if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
5636 !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
5637 kvm->arch.sgx_provisioning_allowed = true;
5643 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
5645 if (kvm_x86_ops.vm_copy_enc_context_from)
5646 r = kvm_x86_ops.vm_copy_enc_context_from(kvm, cap->args[0]);
5648 case KVM_CAP_EXIT_HYPERCALL:
5649 if (cap->args[0] & ~KVM_EXIT_HYPERCALL_VALID_MASK) {
5653 kvm->arch.hypercall_exit_enabled = cap->args[0];
5656 case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
5658 if (cap->args[0] & ~1)
5660 kvm->arch.exit_on_emulation_error = cap->args[0];
5670 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
5672 struct kvm_x86_msr_filter *msr_filter;
5674 msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
5678 msr_filter->default_allow = default_allow;
5682 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
5689 for (i = 0; i < msr_filter->count; i++)
5690 kfree(msr_filter->ranges[i].bitmap);
5695 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
5696 struct kvm_msr_filter_range *user_range)
5698 unsigned long *bitmap = NULL;
5701 if (!user_range->nmsrs)
5704 if (user_range->flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE))
5707 if (!user_range->flags)
5710 bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5711 if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5714 bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5716 return PTR_ERR(bitmap);
5718 msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
5719 .flags = user_range->flags,
5720 .base = user_range->base,
5721 .nmsrs = user_range->nmsrs,
5725 msr_filter->count++;
5729 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5731 struct kvm_msr_filter __user *user_msr_filter = argp;
5732 struct kvm_x86_msr_filter *new_filter, *old_filter;
5733 struct kvm_msr_filter filter;
5739 if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5742 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5743 empty &= !filter.ranges[i].nmsrs;
5745 default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5746 if (empty && !default_allow)
5749 new_filter = kvm_alloc_msr_filter(default_allow);
5753 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5754 r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
5756 kvm_free_msr_filter(new_filter);
5761 mutex_lock(&kvm->lock);
5763 /* The per-VM filter is protected by kvm->lock... */
5764 old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
5766 rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
5767 synchronize_srcu(&kvm->srcu);
5769 kvm_free_msr_filter(old_filter);
5771 kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5772 mutex_unlock(&kvm->lock);
5777 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
5778 static int kvm_arch_suspend_notifier(struct kvm *kvm)
5780 struct kvm_vcpu *vcpu;
5783 mutex_lock(&kvm->lock);
5784 kvm_for_each_vcpu(i, vcpu, kvm) {
5785 if (!vcpu->arch.pv_time_enabled)
5788 ret = kvm_set_guest_paused(vcpu);
5790 kvm_err("Failed to pause guest VCPU%d: %d\n",
5791 vcpu->vcpu_id, ret);
5795 mutex_unlock(&kvm->lock);
5797 return ret ? NOTIFY_BAD : NOTIFY_DONE;
5800 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state)
5803 case PM_HIBERNATION_PREPARE:
5804 case PM_SUSPEND_PREPARE:
5805 return kvm_arch_suspend_notifier(kvm);
5810 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
5812 long kvm_arch_vm_ioctl(struct file *filp,
5813 unsigned int ioctl, unsigned long arg)
5815 struct kvm *kvm = filp->private_data;
5816 void __user *argp = (void __user *)arg;
5819 * This union makes it completely explicit to gcc-3.x
5820 * that these two variables' stack usage should be
5821 * combined, not added together.
5824 struct kvm_pit_state ps;
5825 struct kvm_pit_state2 ps2;
5826 struct kvm_pit_config pit_config;
5830 case KVM_SET_TSS_ADDR:
5831 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5833 case KVM_SET_IDENTITY_MAP_ADDR: {
5836 mutex_lock(&kvm->lock);
5838 if (kvm->created_vcpus)
5839 goto set_identity_unlock;
5841 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5842 goto set_identity_unlock;
5843 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5844 set_identity_unlock:
5845 mutex_unlock(&kvm->lock);
5848 case KVM_SET_NR_MMU_PAGES:
5849 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5851 case KVM_GET_NR_MMU_PAGES:
5852 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5854 case KVM_CREATE_IRQCHIP: {
5855 mutex_lock(&kvm->lock);
5858 if (irqchip_in_kernel(kvm))
5859 goto create_irqchip_unlock;
5862 if (kvm->created_vcpus)
5863 goto create_irqchip_unlock;
5865 r = kvm_pic_init(kvm);
5867 goto create_irqchip_unlock;
5869 r = kvm_ioapic_init(kvm);
5871 kvm_pic_destroy(kvm);
5872 goto create_irqchip_unlock;
5875 r = kvm_setup_default_irq_routing(kvm);
5877 kvm_ioapic_destroy(kvm);
5878 kvm_pic_destroy(kvm);
5879 goto create_irqchip_unlock;
5881 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5883 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5884 create_irqchip_unlock:
5885 mutex_unlock(&kvm->lock);
5888 case KVM_CREATE_PIT:
5889 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5891 case KVM_CREATE_PIT2:
5893 if (copy_from_user(&u.pit_config, argp,
5894 sizeof(struct kvm_pit_config)))
5897 mutex_lock(&kvm->lock);
5900 goto create_pit_unlock;
5902 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5906 mutex_unlock(&kvm->lock);
5908 case KVM_GET_IRQCHIP: {
5909 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5910 struct kvm_irqchip *chip;
5912 chip = memdup_user(argp, sizeof(*chip));
5919 if (!irqchip_kernel(kvm))
5920 goto get_irqchip_out;
5921 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5923 goto get_irqchip_out;
5925 if (copy_to_user(argp, chip, sizeof(*chip)))
5926 goto get_irqchip_out;
5932 case KVM_SET_IRQCHIP: {
5933 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5934 struct kvm_irqchip *chip;
5936 chip = memdup_user(argp, sizeof(*chip));
5943 if (!irqchip_kernel(kvm))
5944 goto set_irqchip_out;
5945 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5952 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5955 if (!kvm->arch.vpit)
5957 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5961 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5968 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5970 mutex_lock(&kvm->lock);
5972 if (!kvm->arch.vpit)
5974 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5976 mutex_unlock(&kvm->lock);
5979 case KVM_GET_PIT2: {
5981 if (!kvm->arch.vpit)
5983 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5987 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5992 case KVM_SET_PIT2: {
5994 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5996 mutex_lock(&kvm->lock);
5998 if (!kvm->arch.vpit)
6000 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
6002 mutex_unlock(&kvm->lock);
6005 case KVM_REINJECT_CONTROL: {
6006 struct kvm_reinject_control control;
6008 if (copy_from_user(&control, argp, sizeof(control)))
6011 if (!kvm->arch.vpit)
6013 r = kvm_vm_ioctl_reinject(kvm, &control);
6016 case KVM_SET_BOOT_CPU_ID:
6018 mutex_lock(&kvm->lock);
6019 if (kvm->created_vcpus)
6022 kvm->arch.bsp_vcpu_id = arg;
6023 mutex_unlock(&kvm->lock);
6025 #ifdef CONFIG_KVM_XEN
6026 case KVM_XEN_HVM_CONFIG: {
6027 struct kvm_xen_hvm_config xhc;
6029 if (copy_from_user(&xhc, argp, sizeof(xhc)))
6031 r = kvm_xen_hvm_config(kvm, &xhc);
6034 case KVM_XEN_HVM_GET_ATTR: {
6035 struct kvm_xen_hvm_attr xha;
6038 if (copy_from_user(&xha, argp, sizeof(xha)))
6040 r = kvm_xen_hvm_get_attr(kvm, &xha);
6041 if (!r && copy_to_user(argp, &xha, sizeof(xha)))
6045 case KVM_XEN_HVM_SET_ATTR: {
6046 struct kvm_xen_hvm_attr xha;
6049 if (copy_from_user(&xha, argp, sizeof(xha)))
6051 r = kvm_xen_hvm_set_attr(kvm, &xha);
6055 case KVM_SET_CLOCK: {
6056 struct kvm_arch *ka = &kvm->arch;
6057 struct kvm_clock_data user_ns;
6061 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
6070 * TODO: userspace has to take care of races with VCPU_RUN, so
6071 * kvm_gen_update_masterclock() can be cut down to locked
6072 * pvclock_update_vm_gtod_copy().
6074 kvm_gen_update_masterclock(kvm);
6077 * This pairs with kvm_guest_time_update(): when masterclock is
6078 * in use, we use master_kernel_ns + kvmclock_offset to set
6079 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
6080 * is slightly ahead) here we risk going negative on unsigned
6081 * 'system_time' when 'user_ns.clock' is very small.
6083 spin_lock_irq(&ka->pvclock_gtod_sync_lock);
6084 if (kvm->arch.use_master_clock)
6085 now_ns = ka->master_kernel_ns;
6087 now_ns = get_kvmclock_base_ns();
6088 ka->kvmclock_offset = user_ns.clock - now_ns;
6089 spin_unlock_irq(&ka->pvclock_gtod_sync_lock);
6091 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
6094 case KVM_GET_CLOCK: {
6095 struct kvm_clock_data user_ns;
6098 now_ns = get_kvmclock_ns(kvm);
6099 user_ns.clock = now_ns;
6100 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
6101 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
6104 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
6109 case KVM_MEMORY_ENCRYPT_OP: {
6111 if (kvm_x86_ops.mem_enc_op)
6112 r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
6115 case KVM_MEMORY_ENCRYPT_REG_REGION: {
6116 struct kvm_enc_region region;
6119 if (copy_from_user(®ion, argp, sizeof(region)))
6123 if (kvm_x86_ops.mem_enc_reg_region)
6124 r = static_call(kvm_x86_mem_enc_reg_region)(kvm, ®ion);
6127 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
6128 struct kvm_enc_region region;
6131 if (copy_from_user(®ion, argp, sizeof(region)))
6135 if (kvm_x86_ops.mem_enc_unreg_region)
6136 r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, ®ion);
6139 case KVM_HYPERV_EVENTFD: {
6140 struct kvm_hyperv_eventfd hvevfd;
6143 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
6145 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
6148 case KVM_SET_PMU_EVENT_FILTER:
6149 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
6151 case KVM_X86_SET_MSR_FILTER:
6152 r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
6161 static void kvm_init_msr_list(void)
6163 struct x86_pmu_capability x86_pmu;
6167 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
6168 "Please update the fixed PMCs in msrs_to_saved_all[]");
6170 perf_get_x86_pmu_capability(&x86_pmu);
6172 num_msrs_to_save = 0;
6173 num_emulated_msrs = 0;
6174 num_msr_based_features = 0;
6176 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
6177 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
6181 * Even MSRs that are valid in the host may not be exposed
6182 * to the guests in some cases.
6184 switch (msrs_to_save_all[i]) {
6185 case MSR_IA32_BNDCFGS:
6186 if (!kvm_mpx_supported())
6190 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
6191 !kvm_cpu_cap_has(X86_FEATURE_RDPID))
6194 case MSR_IA32_UMWAIT_CONTROL:
6195 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
6198 case MSR_IA32_RTIT_CTL:
6199 case MSR_IA32_RTIT_STATUS:
6200 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
6203 case MSR_IA32_RTIT_CR3_MATCH:
6204 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6205 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
6208 case MSR_IA32_RTIT_OUTPUT_BASE:
6209 case MSR_IA32_RTIT_OUTPUT_MASK:
6210 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6211 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
6212 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
6215 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
6216 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6217 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
6218 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
6221 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
6222 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
6223 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6226 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
6227 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
6228 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6235 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
6238 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
6239 if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
6242 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
6245 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
6246 struct kvm_msr_entry msr;
6248 msr.index = msr_based_features_all[i];
6249 if (kvm_get_msr_feature(&msr))
6252 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6256 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6264 if (!(lapic_in_kernel(vcpu) &&
6265 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6266 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6277 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6284 if (!(lapic_in_kernel(vcpu) &&
6285 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6287 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6289 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
6299 static void kvm_set_segment(struct kvm_vcpu *vcpu,
6300 struct kvm_segment *var, int seg)
6302 static_call(kvm_x86_set_segment)(vcpu, var, seg);
6305 void kvm_get_segment(struct kvm_vcpu *vcpu,
6306 struct kvm_segment *var, int seg)
6308 static_call(kvm_x86_get_segment)(vcpu, var, seg);
6311 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
6312 struct x86_exception *exception)
6316 BUG_ON(!mmu_is_nested(vcpu));
6318 /* NPT walks are always user-walks */
6319 access |= PFERR_USER_MASK;
6320 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
6325 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
6326 struct x86_exception *exception)
6328 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6329 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6331 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
6333 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
6334 struct x86_exception *exception)
6336 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6337 access |= PFERR_FETCH_MASK;
6338 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6341 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
6342 struct x86_exception *exception)
6344 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6345 access |= PFERR_WRITE_MASK;
6346 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6348 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
6350 /* uses this to access any guest's mapped memory without checking CPL */
6351 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6352 struct x86_exception *exception)
6354 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
6357 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6358 struct kvm_vcpu *vcpu, u32 access,
6359 struct x86_exception *exception)
6362 int r = X86EMUL_CONTINUE;
6365 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
6367 unsigned offset = addr & (PAGE_SIZE-1);
6368 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6371 if (gpa == UNMAPPED_GVA)
6372 return X86EMUL_PROPAGATE_FAULT;
6373 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6376 r = X86EMUL_IO_NEEDED;
6388 /* used for instruction fetching */
6389 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6390 gva_t addr, void *val, unsigned int bytes,
6391 struct x86_exception *exception)
6393 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6394 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6398 /* Inline kvm_read_guest_virt_helper for speed. */
6399 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
6401 if (unlikely(gpa == UNMAPPED_GVA))
6402 return X86EMUL_PROPAGATE_FAULT;
6404 offset = addr & (PAGE_SIZE-1);
6405 if (WARN_ON(offset + bytes > PAGE_SIZE))
6406 bytes = (unsigned)PAGE_SIZE - offset;
6407 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6409 if (unlikely(ret < 0))
6410 return X86EMUL_IO_NEEDED;
6412 return X86EMUL_CONTINUE;
6415 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6416 gva_t addr, void *val, unsigned int bytes,
6417 struct x86_exception *exception)
6419 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6422 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6423 * is returned, but our callers are not ready for that and they blindly
6424 * call kvm_inject_page_fault. Ensure that they at least do not leak
6425 * uninitialized kernel stack memory into cr2 and error code.
6427 memset(exception, 0, sizeof(*exception));
6428 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6431 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6433 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6434 gva_t addr, void *val, unsigned int bytes,
6435 struct x86_exception *exception, bool system)
6437 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6440 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6441 access |= PFERR_USER_MASK;
6443 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6446 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6447 unsigned long addr, void *val, unsigned int bytes)
6449 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6450 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6452 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6455 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6456 struct kvm_vcpu *vcpu, u32 access,
6457 struct x86_exception *exception)
6460 int r = X86EMUL_CONTINUE;
6463 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6466 unsigned offset = addr & (PAGE_SIZE-1);
6467 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6470 if (gpa == UNMAPPED_GVA)
6471 return X86EMUL_PROPAGATE_FAULT;
6472 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6474 r = X86EMUL_IO_NEEDED;
6486 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6487 unsigned int bytes, struct x86_exception *exception,
6490 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6491 u32 access = PFERR_WRITE_MASK;
6493 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6494 access |= PFERR_USER_MASK;
6496 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6500 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6501 unsigned int bytes, struct x86_exception *exception)
6503 /* kvm_write_guest_virt_system can pull in tons of pages. */
6504 vcpu->arch.l1tf_flush_l1d = true;
6506 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6507 PFERR_WRITE_MASK, exception);
6509 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6511 int handle_ud(struct kvm_vcpu *vcpu)
6513 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6514 int emul_type = EMULTYPE_TRAP_UD;
6515 char sig[5]; /* ud2; .ascii "kvm" */
6516 struct x86_exception e;
6518 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6521 if (force_emulation_prefix &&
6522 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6523 sig, sizeof(sig), &e) == 0 &&
6524 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6525 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6526 emul_type = EMULTYPE_TRAP_UD_FORCED;
6529 return kvm_emulate_instruction(vcpu, emul_type);
6531 EXPORT_SYMBOL_GPL(handle_ud);
6533 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6534 gpa_t gpa, bool write)
6536 /* For APIC access vmexit */
6537 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6540 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6541 trace_vcpu_match_mmio(gva, gpa, write, true);
6548 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6549 gpa_t *gpa, struct x86_exception *exception,
6552 u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6553 | (write ? PFERR_WRITE_MASK : 0);
6556 * currently PKRU is only applied to ept enabled guest so
6557 * there is no pkey in EPT page table for L1 guest or EPT
6558 * shadow page table for L2 guest.
6560 if (vcpu_match_mmio_gva(vcpu, gva) && (!is_paging(vcpu) ||
6561 !permission_fault(vcpu, vcpu->arch.walk_mmu,
6562 vcpu->arch.mmio_access, 0, access))) {
6563 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6564 (gva & (PAGE_SIZE - 1));
6565 trace_vcpu_match_mmio(gva, *gpa, write, false);
6569 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6571 if (*gpa == UNMAPPED_GVA)
6574 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6577 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6578 const void *val, int bytes)
6582 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6585 kvm_page_track_write(vcpu, gpa, val, bytes);
6589 struct read_write_emulator_ops {
6590 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6592 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6593 void *val, int bytes);
6594 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6595 int bytes, void *val);
6596 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6597 void *val, int bytes);
6601 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6603 if (vcpu->mmio_read_completed) {
6604 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6605 vcpu->mmio_fragments[0].gpa, val);
6606 vcpu->mmio_read_completed = 0;
6613 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6614 void *val, int bytes)
6616 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6619 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6620 void *val, int bytes)
6622 return emulator_write_phys(vcpu, gpa, val, bytes);
6625 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6627 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6628 return vcpu_mmio_write(vcpu, gpa, bytes, val);
6631 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6632 void *val, int bytes)
6634 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6635 return X86EMUL_IO_NEEDED;
6638 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6639 void *val, int bytes)
6641 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6643 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6644 return X86EMUL_CONTINUE;
6647 static const struct read_write_emulator_ops read_emultor = {
6648 .read_write_prepare = read_prepare,
6649 .read_write_emulate = read_emulate,
6650 .read_write_mmio = vcpu_mmio_read,
6651 .read_write_exit_mmio = read_exit_mmio,
6654 static const struct read_write_emulator_ops write_emultor = {
6655 .read_write_emulate = write_emulate,
6656 .read_write_mmio = write_mmio,
6657 .read_write_exit_mmio = write_exit_mmio,
6661 static int emulator_read_write_onepage(unsigned long addr, void *val,
6663 struct x86_exception *exception,
6664 struct kvm_vcpu *vcpu,
6665 const struct read_write_emulator_ops *ops)
6669 bool write = ops->write;
6670 struct kvm_mmio_fragment *frag;
6671 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6674 * If the exit was due to a NPF we may already have a GPA.
6675 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6676 * Note, this cannot be used on string operations since string
6677 * operation using rep will only have the initial GPA from the NPF
6680 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6681 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6682 gpa = ctxt->gpa_val;
6683 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6685 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6687 return X86EMUL_PROPAGATE_FAULT;
6690 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6691 return X86EMUL_CONTINUE;
6694 * Is this MMIO handled locally?
6696 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6697 if (handled == bytes)
6698 return X86EMUL_CONTINUE;
6704 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6705 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6709 return X86EMUL_CONTINUE;
6712 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6714 void *val, unsigned int bytes,
6715 struct x86_exception *exception,
6716 const struct read_write_emulator_ops *ops)
6718 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6722 if (ops->read_write_prepare &&
6723 ops->read_write_prepare(vcpu, val, bytes))
6724 return X86EMUL_CONTINUE;
6726 vcpu->mmio_nr_fragments = 0;
6728 /* Crossing a page boundary? */
6729 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6732 now = -addr & ~PAGE_MASK;
6733 rc = emulator_read_write_onepage(addr, val, now, exception,
6736 if (rc != X86EMUL_CONTINUE)
6739 if (ctxt->mode != X86EMUL_MODE_PROT64)
6745 rc = emulator_read_write_onepage(addr, val, bytes, exception,
6747 if (rc != X86EMUL_CONTINUE)
6750 if (!vcpu->mmio_nr_fragments)
6753 gpa = vcpu->mmio_fragments[0].gpa;
6755 vcpu->mmio_needed = 1;
6756 vcpu->mmio_cur_fragment = 0;
6758 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6759 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6760 vcpu->run->exit_reason = KVM_EXIT_MMIO;
6761 vcpu->run->mmio.phys_addr = gpa;
6763 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6766 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6770 struct x86_exception *exception)
6772 return emulator_read_write(ctxt, addr, val, bytes,
6773 exception, &read_emultor);
6776 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6780 struct x86_exception *exception)
6782 return emulator_read_write(ctxt, addr, (void *)val, bytes,
6783 exception, &write_emultor);
6786 #define CMPXCHG_TYPE(t, ptr, old, new) \
6787 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6789 #ifdef CONFIG_X86_64
6790 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6792 # define CMPXCHG64(ptr, old, new) \
6793 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6796 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6801 struct x86_exception *exception)
6803 struct kvm_host_map map;
6804 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6810 /* guests cmpxchg8b have to be emulated atomically */
6811 if (bytes > 8 || (bytes & (bytes - 1)))
6814 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6816 if (gpa == UNMAPPED_GVA ||
6817 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6821 * Emulate the atomic as a straight write to avoid #AC if SLD is
6822 * enabled in the host and the access splits a cache line.
6824 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6825 page_line_mask = ~(cache_line_size() - 1);
6827 page_line_mask = PAGE_MASK;
6829 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6832 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6835 kaddr = map.hva + offset_in_page(gpa);
6839 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6842 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6845 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6848 exchanged = CMPXCHG64(kaddr, old, new);
6854 kvm_vcpu_unmap(vcpu, &map, true);
6857 return X86EMUL_CMPXCHG_FAILED;
6859 kvm_page_track_write(vcpu, gpa, new, bytes);
6861 return X86EMUL_CONTINUE;
6864 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6866 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6869 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6873 for (i = 0; i < vcpu->arch.pio.count; i++) {
6874 if (vcpu->arch.pio.in)
6875 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6876 vcpu->arch.pio.size, pd);
6878 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6879 vcpu->arch.pio.port, vcpu->arch.pio.size,
6883 pd += vcpu->arch.pio.size;
6888 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6889 unsigned short port, void *val,
6890 unsigned int count, bool in)
6892 vcpu->arch.pio.port = port;
6893 vcpu->arch.pio.in = in;
6894 vcpu->arch.pio.count = count;
6895 vcpu->arch.pio.size = size;
6897 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6898 vcpu->arch.pio.count = 0;
6902 vcpu->run->exit_reason = KVM_EXIT_IO;
6903 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6904 vcpu->run->io.size = size;
6905 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6906 vcpu->run->io.count = count;
6907 vcpu->run->io.port = port;
6912 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6913 unsigned short port, void *val, unsigned int count)
6917 if (vcpu->arch.pio.count)
6920 memset(vcpu->arch.pio_data, 0, size * count);
6922 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6925 memcpy(val, vcpu->arch.pio_data, size * count);
6926 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6927 vcpu->arch.pio.count = 0;
6934 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6935 int size, unsigned short port, void *val,
6938 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6942 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6943 unsigned short port, const void *val,
6946 memcpy(vcpu->arch.pio_data, val, size * count);
6947 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6948 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6951 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6952 int size, unsigned short port,
6953 const void *val, unsigned int count)
6955 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6958 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6960 return static_call(kvm_x86_get_segment_base)(vcpu, seg);
6963 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6965 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6968 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6970 if (!need_emulate_wbinvd(vcpu))
6971 return X86EMUL_CONTINUE;
6973 if (static_call(kvm_x86_has_wbinvd_exit)()) {
6974 int cpu = get_cpu();
6976 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6977 on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
6978 wbinvd_ipi, NULL, 1);
6980 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6983 return X86EMUL_CONTINUE;
6986 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6988 kvm_emulate_wbinvd_noskip(vcpu);
6989 return kvm_skip_emulated_instruction(vcpu);
6991 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6995 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6997 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
7000 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
7001 unsigned long *dest)
7003 kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
7006 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
7007 unsigned long value)
7010 return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
7013 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
7015 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
7018 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
7020 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7021 unsigned long value;
7025 value = kvm_read_cr0(vcpu);
7028 value = vcpu->arch.cr2;
7031 value = kvm_read_cr3(vcpu);
7034 value = kvm_read_cr4(vcpu);
7037 value = kvm_get_cr8(vcpu);
7040 kvm_err("%s: unexpected cr %u\n", __func__, cr);
7047 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
7049 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7054 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
7057 vcpu->arch.cr2 = val;
7060 res = kvm_set_cr3(vcpu, val);
7063 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
7066 res = kvm_set_cr8(vcpu, val);
7069 kvm_err("%s: unexpected cr %u\n", __func__, cr);
7076 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
7078 return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
7081 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7083 static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
7086 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7088 static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
7091 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7093 static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
7096 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7098 static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
7101 static unsigned long emulator_get_cached_segment_base(
7102 struct x86_emulate_ctxt *ctxt, int seg)
7104 return get_segment_base(emul_to_vcpu(ctxt), seg);
7107 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
7108 struct desc_struct *desc, u32 *base3,
7111 struct kvm_segment var;
7113 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
7114 *selector = var.selector;
7117 memset(desc, 0, sizeof(*desc));
7125 set_desc_limit(desc, var.limit);
7126 set_desc_base(desc, (unsigned long)var.base);
7127 #ifdef CONFIG_X86_64
7129 *base3 = var.base >> 32;
7131 desc->type = var.type;
7133 desc->dpl = var.dpl;
7134 desc->p = var.present;
7135 desc->avl = var.avl;
7143 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
7144 struct desc_struct *desc, u32 base3,
7147 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7148 struct kvm_segment var;
7150 var.selector = selector;
7151 var.base = get_desc_base(desc);
7152 #ifdef CONFIG_X86_64
7153 var.base |= ((u64)base3) << 32;
7155 var.limit = get_desc_limit(desc);
7157 var.limit = (var.limit << 12) | 0xfff;
7158 var.type = desc->type;
7159 var.dpl = desc->dpl;
7164 var.avl = desc->avl;
7165 var.present = desc->p;
7166 var.unusable = !var.present;
7169 kvm_set_segment(vcpu, &var, seg);
7173 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
7174 u32 msr_index, u64 *pdata)
7176 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7179 r = kvm_get_msr(vcpu, msr_index, pdata);
7181 if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
7182 /* Bounce to user space */
7183 return X86EMUL_IO_NEEDED;
7189 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
7190 u32 msr_index, u64 data)
7192 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7195 r = kvm_set_msr(vcpu, msr_index, data);
7197 if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
7198 /* Bounce to user space */
7199 return X86EMUL_IO_NEEDED;
7205 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
7207 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7209 return vcpu->arch.smbase;
7212 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
7214 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7216 vcpu->arch.smbase = smbase;
7219 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
7222 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
7225 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
7226 u32 pmc, u64 *pdata)
7228 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
7231 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
7233 emul_to_vcpu(ctxt)->arch.halt_request = 1;
7236 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
7237 struct x86_instruction_info *info,
7238 enum x86_intercept_stage stage)
7240 return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
7244 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
7245 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
7248 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
7251 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
7253 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
7256 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
7258 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
7261 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
7263 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
7266 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
7268 return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
7271 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
7273 kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
7276 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
7278 static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
7281 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
7283 return emul_to_vcpu(ctxt)->arch.hflags;
7286 static void emulator_exiting_smm(struct x86_emulate_ctxt *ctxt)
7288 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7290 kvm_smm_changed(vcpu, false);
7293 static int emulator_leave_smm(struct x86_emulate_ctxt *ctxt,
7294 const char *smstate)
7296 return static_call(kvm_x86_leave_smm)(emul_to_vcpu(ctxt), smstate);
7299 static void emulator_triple_fault(struct x86_emulate_ctxt *ctxt)
7301 kvm_make_request(KVM_REQ_TRIPLE_FAULT, emul_to_vcpu(ctxt));
7304 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
7306 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
7309 static const struct x86_emulate_ops emulate_ops = {
7310 .read_gpr = emulator_read_gpr,
7311 .write_gpr = emulator_write_gpr,
7312 .read_std = emulator_read_std,
7313 .write_std = emulator_write_std,
7314 .read_phys = kvm_read_guest_phys_system,
7315 .fetch = kvm_fetch_guest_virt,
7316 .read_emulated = emulator_read_emulated,
7317 .write_emulated = emulator_write_emulated,
7318 .cmpxchg_emulated = emulator_cmpxchg_emulated,
7319 .invlpg = emulator_invlpg,
7320 .pio_in_emulated = emulator_pio_in_emulated,
7321 .pio_out_emulated = emulator_pio_out_emulated,
7322 .get_segment = emulator_get_segment,
7323 .set_segment = emulator_set_segment,
7324 .get_cached_segment_base = emulator_get_cached_segment_base,
7325 .get_gdt = emulator_get_gdt,
7326 .get_idt = emulator_get_idt,
7327 .set_gdt = emulator_set_gdt,
7328 .set_idt = emulator_set_idt,
7329 .get_cr = emulator_get_cr,
7330 .set_cr = emulator_set_cr,
7331 .cpl = emulator_get_cpl,
7332 .get_dr = emulator_get_dr,
7333 .set_dr = emulator_set_dr,
7334 .get_smbase = emulator_get_smbase,
7335 .set_smbase = emulator_set_smbase,
7336 .set_msr = emulator_set_msr,
7337 .get_msr = emulator_get_msr,
7338 .check_pmc = emulator_check_pmc,
7339 .read_pmc = emulator_read_pmc,
7340 .halt = emulator_halt,
7341 .wbinvd = emulator_wbinvd,
7342 .fix_hypercall = emulator_fix_hypercall,
7343 .intercept = emulator_intercept,
7344 .get_cpuid = emulator_get_cpuid,
7345 .guest_has_long_mode = emulator_guest_has_long_mode,
7346 .guest_has_movbe = emulator_guest_has_movbe,
7347 .guest_has_fxsr = emulator_guest_has_fxsr,
7348 .set_nmi_mask = emulator_set_nmi_mask,
7349 .get_hflags = emulator_get_hflags,
7350 .exiting_smm = emulator_exiting_smm,
7351 .leave_smm = emulator_leave_smm,
7352 .triple_fault = emulator_triple_fault,
7353 .set_xcr = emulator_set_xcr,
7356 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7358 u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7360 * an sti; sti; sequence only disable interrupts for the first
7361 * instruction. So, if the last instruction, be it emulated or
7362 * not, left the system with the INT_STI flag enabled, it
7363 * means that the last instruction is an sti. We should not
7364 * leave the flag on in this case. The same goes for mov ss
7366 if (int_shadow & mask)
7368 if (unlikely(int_shadow || mask)) {
7369 static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7371 kvm_make_request(KVM_REQ_EVENT, vcpu);
7375 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7377 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7378 if (ctxt->exception.vector == PF_VECTOR)
7379 return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7381 if (ctxt->exception.error_code_valid)
7382 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7383 ctxt->exception.error_code);
7385 kvm_queue_exception(vcpu, ctxt->exception.vector);
7389 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7391 struct x86_emulate_ctxt *ctxt;
7393 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7395 pr_err("kvm: failed to allocate vcpu's emulator\n");
7400 ctxt->ops = &emulate_ops;
7401 vcpu->arch.emulate_ctxt = ctxt;
7406 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7408 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7411 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7413 ctxt->gpa_available = false;
7414 ctxt->eflags = kvm_get_rflags(vcpu);
7415 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7417 ctxt->eip = kvm_rip_read(vcpu);
7418 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
7419 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
7420 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
7421 cs_db ? X86EMUL_MODE_PROT32 :
7422 X86EMUL_MODE_PROT16;
7423 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7424 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7425 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7427 ctxt->interruptibility = 0;
7428 ctxt->have_exception = false;
7429 ctxt->exception.vector = -1;
7430 ctxt->perm_ok = false;
7432 init_decode_cache(ctxt);
7433 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7436 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7438 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7441 init_emulate_ctxt(vcpu);
7445 ctxt->_eip = ctxt->eip + inc_eip;
7446 ret = emulate_int_real(ctxt, irq);
7448 if (ret != X86EMUL_CONTINUE) {
7449 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7451 ctxt->eip = ctxt->_eip;
7452 kvm_rip_write(vcpu, ctxt->eip);
7453 kvm_set_rflags(vcpu, ctxt->eflags);
7456 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7458 static void prepare_emulation_failure_exit(struct kvm_vcpu *vcpu)
7460 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7461 u32 insn_size = ctxt->fetch.end - ctxt->fetch.data;
7462 struct kvm_run *run = vcpu->run;
7464 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7465 run->emulation_failure.suberror = KVM_INTERNAL_ERROR_EMULATION;
7466 run->emulation_failure.ndata = 0;
7467 run->emulation_failure.flags = 0;
7470 run->emulation_failure.ndata = 3;
7471 run->emulation_failure.flags |=
7472 KVM_INTERNAL_ERROR_EMULATION_FLAG_INSTRUCTION_BYTES;
7473 run->emulation_failure.insn_size = insn_size;
7474 memset(run->emulation_failure.insn_bytes, 0x90,
7475 sizeof(run->emulation_failure.insn_bytes));
7476 memcpy(run->emulation_failure.insn_bytes,
7477 ctxt->fetch.data, insn_size);
7481 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7483 struct kvm *kvm = vcpu->kvm;
7485 ++vcpu->stat.insn_emulation_fail;
7486 trace_kvm_emulate_insn_failed(vcpu);
7488 if (emulation_type & EMULTYPE_VMWARE_GP) {
7489 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7493 if (kvm->arch.exit_on_emulation_error ||
7494 (emulation_type & EMULTYPE_SKIP)) {
7495 prepare_emulation_failure_exit(vcpu);
7499 kvm_queue_exception(vcpu, UD_VECTOR);
7501 if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7502 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7503 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7504 vcpu->run->internal.ndata = 0;
7511 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7512 bool write_fault_to_shadow_pgtable,
7515 gpa_t gpa = cr2_or_gpa;
7518 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7521 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7522 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7525 if (!vcpu->arch.mmu->direct_map) {
7527 * Write permission should be allowed since only
7528 * write access need to be emulated.
7530 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7533 * If the mapping is invalid in guest, let cpu retry
7534 * it to generate fault.
7536 if (gpa == UNMAPPED_GVA)
7541 * Do not retry the unhandleable instruction if it faults on the
7542 * readonly host memory, otherwise it will goto a infinite loop:
7543 * retry instruction -> write #PF -> emulation fail -> retry
7544 * instruction -> ...
7546 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7549 * If the instruction failed on the error pfn, it can not be fixed,
7550 * report the error to userspace.
7552 if (is_error_noslot_pfn(pfn))
7555 kvm_release_pfn_clean(pfn);
7557 /* The instructions are well-emulated on direct mmu. */
7558 if (vcpu->arch.mmu->direct_map) {
7559 unsigned int indirect_shadow_pages;
7561 write_lock(&vcpu->kvm->mmu_lock);
7562 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7563 write_unlock(&vcpu->kvm->mmu_lock);
7565 if (indirect_shadow_pages)
7566 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7572 * if emulation was due to access to shadowed page table
7573 * and it failed try to unshadow page and re-enter the
7574 * guest to let CPU execute the instruction.
7576 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7579 * If the access faults on its page table, it can not
7580 * be fixed by unprotecting shadow page and it should
7581 * be reported to userspace.
7583 return !write_fault_to_shadow_pgtable;
7586 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7587 gpa_t cr2_or_gpa, int emulation_type)
7589 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7590 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7592 last_retry_eip = vcpu->arch.last_retry_eip;
7593 last_retry_addr = vcpu->arch.last_retry_addr;
7596 * If the emulation is caused by #PF and it is non-page_table
7597 * writing instruction, it means the VM-EXIT is caused by shadow
7598 * page protected, we can zap the shadow page and retry this
7599 * instruction directly.
7601 * Note: if the guest uses a non-page-table modifying instruction
7602 * on the PDE that points to the instruction, then we will unmap
7603 * the instruction and go to an infinite loop. So, we cache the
7604 * last retried eip and the last fault address, if we meet the eip
7605 * and the address again, we can break out of the potential infinite
7608 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7610 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7613 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7614 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7617 if (x86_page_table_writing_insn(ctxt))
7620 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7623 vcpu->arch.last_retry_eip = ctxt->eip;
7624 vcpu->arch.last_retry_addr = cr2_or_gpa;
7626 if (!vcpu->arch.mmu->direct_map)
7627 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7629 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7634 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7635 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7637 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm)
7639 trace_kvm_smm_transition(vcpu->vcpu_id, vcpu->arch.smbase, entering_smm);
7642 vcpu->arch.hflags |= HF_SMM_MASK;
7644 vcpu->arch.hflags &= ~(HF_SMM_MASK | HF_SMM_INSIDE_NMI_MASK);
7646 /* Process a latched INIT or SMI, if any. */
7647 kvm_make_request(KVM_REQ_EVENT, vcpu);
7650 * Even if KVM_SET_SREGS2 loaded PDPTRs out of band,
7651 * on SMM exit we still need to reload them from
7654 vcpu->arch.pdptrs_from_userspace = false;
7657 kvm_mmu_reset_context(vcpu);
7660 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7669 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7670 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7675 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7677 struct kvm_run *kvm_run = vcpu->run;
7679 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7680 kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
7681 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7682 kvm_run->debug.arch.exception = DB_VECTOR;
7683 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7686 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7690 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7692 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7695 r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
7700 * rflags is the old, "raw" value of the flags. The new value has
7701 * not been saved yet.
7703 * This is correct even for TF set by the guest, because "the
7704 * processor will not generate this exception after the instruction
7705 * that sets the TF flag".
7707 if (unlikely(rflags & X86_EFLAGS_TF))
7708 r = kvm_vcpu_do_singlestep(vcpu);
7711 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7713 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7715 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7716 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7717 struct kvm_run *kvm_run = vcpu->run;
7718 unsigned long eip = kvm_get_linear_rip(vcpu);
7719 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7720 vcpu->arch.guest_debug_dr7,
7724 kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
7725 kvm_run->debug.arch.pc = eip;
7726 kvm_run->debug.arch.exception = DB_VECTOR;
7727 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7733 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7734 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7735 unsigned long eip = kvm_get_linear_rip(vcpu);
7736 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7741 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7750 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7752 switch (ctxt->opcode_len) {
7759 case 0xe6: /* OUT */
7763 case 0x6c: /* INS */
7765 case 0x6e: /* OUTS */
7772 case 0x33: /* RDPMC */
7782 * Decode to be emulated instruction. Return EMULATION_OK if success.
7784 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
7785 void *insn, int insn_len)
7787 int r = EMULATION_OK;
7788 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7790 init_emulate_ctxt(vcpu);
7793 * We will reenter on the same instruction since we do not set
7794 * complete_userspace_io. This does not handle watchpoints yet,
7795 * those would be handled in the emulate_ops.
7797 if (!(emulation_type & EMULTYPE_SKIP) &&
7798 kvm_vcpu_check_breakpoint(vcpu, &r))
7801 r = x86_decode_insn(ctxt, insn, insn_len, emulation_type);
7803 trace_kvm_emulate_insn_start(vcpu);
7804 ++vcpu->stat.insn_emulation;
7808 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
7810 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7811 int emulation_type, void *insn, int insn_len)
7814 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7815 bool writeback = true;
7816 bool write_fault_to_spt;
7818 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
7821 vcpu->arch.l1tf_flush_l1d = true;
7824 * Clear write_fault_to_shadow_pgtable here to ensure it is
7827 write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7828 vcpu->arch.write_fault_to_shadow_pgtable = false;
7830 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7831 kvm_clear_exception_queue(vcpu);
7833 r = x86_decode_emulated_instruction(vcpu, emulation_type,
7835 if (r != EMULATION_OK) {
7836 if ((emulation_type & EMULTYPE_TRAP_UD) ||
7837 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7838 kvm_queue_exception(vcpu, UD_VECTOR);
7841 if (reexecute_instruction(vcpu, cr2_or_gpa,
7845 if (ctxt->have_exception) {
7847 * #UD should result in just EMULATION_FAILED, and trap-like
7848 * exception should not be encountered during decode.
7850 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
7851 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
7852 inject_emulated_exception(vcpu);
7855 return handle_emulation_failure(vcpu, emulation_type);
7859 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
7860 !is_vmware_backdoor_opcode(ctxt)) {
7861 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7866 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7867 * for kvm_skip_emulated_instruction(). The caller is responsible for
7868 * updating interruptibility state and injecting single-step #DBs.
7870 if (emulation_type & EMULTYPE_SKIP) {
7871 kvm_rip_write(vcpu, ctxt->_eip);
7872 if (ctxt->eflags & X86_EFLAGS_RF)
7873 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
7877 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
7880 /* this is needed for vmware backdoor interface to work since it
7881 changes registers values during IO operation */
7882 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
7883 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7884 emulator_invalidate_register_cache(ctxt);
7888 if (emulation_type & EMULTYPE_PF) {
7889 /* Save the faulting GPA (cr2) in the address field */
7890 ctxt->exception.address = cr2_or_gpa;
7892 /* With shadow page tables, cr2 contains a GVA or nGPA. */
7893 if (vcpu->arch.mmu->direct_map) {
7894 ctxt->gpa_available = true;
7895 ctxt->gpa_val = cr2_or_gpa;
7898 /* Sanitize the address out of an abundance of paranoia. */
7899 ctxt->exception.address = 0;
7902 r = x86_emulate_insn(ctxt);
7904 if (r == EMULATION_INTERCEPTED)
7907 if (r == EMULATION_FAILED) {
7908 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7912 return handle_emulation_failure(vcpu, emulation_type);
7915 if (ctxt->have_exception) {
7917 if (inject_emulated_exception(vcpu))
7919 } else if (vcpu->arch.pio.count) {
7920 if (!vcpu->arch.pio.in) {
7921 /* FIXME: return into emulator if single-stepping. */
7922 vcpu->arch.pio.count = 0;
7925 vcpu->arch.complete_userspace_io = complete_emulated_pio;
7928 } else if (vcpu->mmio_needed) {
7929 ++vcpu->stat.mmio_exits;
7931 if (!vcpu->mmio_is_write)
7934 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7935 } else if (r == EMULATION_RESTART)
7941 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7942 toggle_interruptibility(vcpu, ctxt->interruptibility);
7943 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7944 if (!ctxt->have_exception ||
7945 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7946 kvm_rip_write(vcpu, ctxt->eip);
7947 if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7948 r = kvm_vcpu_do_singlestep(vcpu);
7949 if (kvm_x86_ops.update_emulated_instruction)
7950 static_call(kvm_x86_update_emulated_instruction)(vcpu);
7951 __kvm_set_rflags(vcpu, ctxt->eflags);
7955 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7956 * do nothing, and it will be requested again as soon as
7957 * the shadow expires. But we still need to check here,
7958 * because POPF has no interrupt shadow.
7960 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7961 kvm_make_request(KVM_REQ_EVENT, vcpu);
7963 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7968 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7970 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7972 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7974 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7975 void *insn, int insn_len)
7977 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7979 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7981 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7983 vcpu->arch.pio.count = 0;
7987 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7989 vcpu->arch.pio.count = 0;
7991 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7994 return kvm_skip_emulated_instruction(vcpu);
7997 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7998 unsigned short port)
8000 unsigned long val = kvm_rax_read(vcpu);
8001 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
8007 * Workaround userspace that relies on old KVM behavior of %rip being
8008 * incremented prior to exiting to userspace to handle "OUT 0x7e".
8011 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
8012 vcpu->arch.complete_userspace_io =
8013 complete_fast_pio_out_port_0x7e;
8014 kvm_skip_emulated_instruction(vcpu);
8016 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8017 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
8022 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
8026 /* We should only ever be called with arch.pio.count equal to 1 */
8027 BUG_ON(vcpu->arch.pio.count != 1);
8029 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
8030 vcpu->arch.pio.count = 0;
8034 /* For size less than 4 we merge, else we zero extend */
8035 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
8038 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
8039 * the copy and tracing
8041 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
8042 kvm_rax_write(vcpu, val);
8044 return kvm_skip_emulated_instruction(vcpu);
8047 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
8048 unsigned short port)
8053 /* For size less than 4 we merge, else we zero extend */
8054 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
8056 ret = emulator_pio_in(vcpu, size, port, &val, 1);
8058 kvm_rax_write(vcpu, val);
8062 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8063 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
8068 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
8073 ret = kvm_fast_pio_in(vcpu, size, port);
8075 ret = kvm_fast_pio_out(vcpu, size, port);
8076 return ret && kvm_skip_emulated_instruction(vcpu);
8078 EXPORT_SYMBOL_GPL(kvm_fast_pio);
8080 static int kvmclock_cpu_down_prep(unsigned int cpu)
8082 __this_cpu_write(cpu_tsc_khz, 0);
8086 static void tsc_khz_changed(void *data)
8088 struct cpufreq_freqs *freq = data;
8089 unsigned long khz = 0;
8093 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8094 khz = cpufreq_quick_get(raw_smp_processor_id());
8097 __this_cpu_write(cpu_tsc_khz, khz);
8100 #ifdef CONFIG_X86_64
8101 static void kvm_hyperv_tsc_notifier(void)
8104 struct kvm_vcpu *vcpu;
8106 unsigned long flags;
8108 mutex_lock(&kvm_lock);
8109 list_for_each_entry(kvm, &vm_list, vm_list)
8110 kvm_make_mclock_inprogress_request(kvm);
8112 hyperv_stop_tsc_emulation();
8114 /* TSC frequency always matches when on Hyper-V */
8115 for_each_present_cpu(cpu)
8116 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
8117 kvm_max_guest_tsc_khz = tsc_khz;
8119 list_for_each_entry(kvm, &vm_list, vm_list) {
8120 struct kvm_arch *ka = &kvm->arch;
8122 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
8123 pvclock_update_vm_gtod_copy(kvm);
8124 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
8126 kvm_for_each_vcpu(cpu, vcpu, kvm)
8127 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8129 kvm_for_each_vcpu(cpu, vcpu, kvm)
8130 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
8132 mutex_unlock(&kvm_lock);
8136 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
8139 struct kvm_vcpu *vcpu;
8140 int i, send_ipi = 0;
8143 * We allow guests to temporarily run on slowing clocks,
8144 * provided we notify them after, or to run on accelerating
8145 * clocks, provided we notify them before. Thus time never
8148 * However, we have a problem. We can't atomically update
8149 * the frequency of a given CPU from this function; it is
8150 * merely a notifier, which can be called from any CPU.
8151 * Changing the TSC frequency at arbitrary points in time
8152 * requires a recomputation of local variables related to
8153 * the TSC for each VCPU. We must flag these local variables
8154 * to be updated and be sure the update takes place with the
8155 * new frequency before any guests proceed.
8157 * Unfortunately, the combination of hotplug CPU and frequency
8158 * change creates an intractable locking scenario; the order
8159 * of when these callouts happen is undefined with respect to
8160 * CPU hotplug, and they can race with each other. As such,
8161 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
8162 * undefined; you can actually have a CPU frequency change take
8163 * place in between the computation of X and the setting of the
8164 * variable. To protect against this problem, all updates of
8165 * the per_cpu tsc_khz variable are done in an interrupt
8166 * protected IPI, and all callers wishing to update the value
8167 * must wait for a synchronous IPI to complete (which is trivial
8168 * if the caller is on the CPU already). This establishes the
8169 * necessary total order on variable updates.
8171 * Note that because a guest time update may take place
8172 * anytime after the setting of the VCPU's request bit, the
8173 * correct TSC value must be set before the request. However,
8174 * to ensure the update actually makes it to any guest which
8175 * starts running in hardware virtualization between the set
8176 * and the acquisition of the spinlock, we must also ping the
8177 * CPU after setting the request bit.
8181 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8183 mutex_lock(&kvm_lock);
8184 list_for_each_entry(kvm, &vm_list, vm_list) {
8185 kvm_for_each_vcpu(i, vcpu, kvm) {
8186 if (vcpu->cpu != cpu)
8188 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8189 if (vcpu->cpu != raw_smp_processor_id())
8193 mutex_unlock(&kvm_lock);
8195 if (freq->old < freq->new && send_ipi) {
8197 * We upscale the frequency. Must make the guest
8198 * doesn't see old kvmclock values while running with
8199 * the new frequency, otherwise we risk the guest sees
8200 * time go backwards.
8202 * In case we update the frequency for another cpu
8203 * (which might be in guest context) send an interrupt
8204 * to kick the cpu out of guest context. Next time
8205 * guest context is entered kvmclock will be updated,
8206 * so the guest will not see stale values.
8208 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8212 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
8215 struct cpufreq_freqs *freq = data;
8218 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
8220 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
8223 for_each_cpu(cpu, freq->policy->cpus)
8224 __kvmclock_cpufreq_notifier(freq, cpu);
8229 static struct notifier_block kvmclock_cpufreq_notifier_block = {
8230 .notifier_call = kvmclock_cpufreq_notifier
8233 static int kvmclock_cpu_online(unsigned int cpu)
8235 tsc_khz_changed(NULL);
8239 static void kvm_timer_init(void)
8241 max_tsc_khz = tsc_khz;
8243 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
8244 #ifdef CONFIG_CPU_FREQ
8245 struct cpufreq_policy *policy;
8249 policy = cpufreq_cpu_get(cpu);
8251 if (policy->cpuinfo.max_freq)
8252 max_tsc_khz = policy->cpuinfo.max_freq;
8253 cpufreq_cpu_put(policy);
8257 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
8258 CPUFREQ_TRANSITION_NOTIFIER);
8261 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
8262 kvmclock_cpu_online, kvmclock_cpu_down_prep);
8265 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
8266 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
8268 int kvm_is_in_guest(void)
8270 return __this_cpu_read(current_vcpu) != NULL;
8273 static int kvm_is_user_mode(void)
8277 if (__this_cpu_read(current_vcpu))
8278 user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu));
8280 return user_mode != 0;
8283 static unsigned long kvm_get_guest_ip(void)
8285 unsigned long ip = 0;
8287 if (__this_cpu_read(current_vcpu))
8288 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
8293 static void kvm_handle_intel_pt_intr(void)
8295 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
8297 kvm_make_request(KVM_REQ_PMI, vcpu);
8298 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8299 (unsigned long *)&vcpu->arch.pmu.global_status);
8302 static struct perf_guest_info_callbacks kvm_guest_cbs = {
8303 .is_in_guest = kvm_is_in_guest,
8304 .is_user_mode = kvm_is_user_mode,
8305 .get_guest_ip = kvm_get_guest_ip,
8306 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
8309 #ifdef CONFIG_X86_64
8310 static void pvclock_gtod_update_fn(struct work_struct *work)
8314 struct kvm_vcpu *vcpu;
8317 mutex_lock(&kvm_lock);
8318 list_for_each_entry(kvm, &vm_list, vm_list)
8319 kvm_for_each_vcpu(i, vcpu, kvm)
8320 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8321 atomic_set(&kvm_guest_has_master_clock, 0);
8322 mutex_unlock(&kvm_lock);
8325 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
8328 * Indirection to move queue_work() out of the tk_core.seq write held
8329 * region to prevent possible deadlocks against time accessors which
8330 * are invoked with work related locks held.
8332 static void pvclock_irq_work_fn(struct irq_work *w)
8334 queue_work(system_long_wq, &pvclock_gtod_work);
8337 static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn);
8340 * Notification about pvclock gtod data update.
8342 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
8345 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
8346 struct timekeeper *tk = priv;
8348 update_pvclock_gtod(tk);
8351 * Disable master clock if host does not trust, or does not use,
8352 * TSC based clocksource. Delegate queue_work() to irq_work as
8353 * this is invoked with tk_core.seq write held.
8355 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
8356 atomic_read(&kvm_guest_has_master_clock) != 0)
8357 irq_work_queue(&pvclock_irq_work);
8361 static struct notifier_block pvclock_gtod_notifier = {
8362 .notifier_call = pvclock_gtod_notify,
8366 int kvm_arch_init(void *opaque)
8368 struct kvm_x86_init_ops *ops = opaque;
8371 if (kvm_x86_ops.hardware_enable) {
8372 printk(KERN_ERR "kvm: already loaded the other module\n");
8377 if (!ops->cpu_has_kvm_support()) {
8378 pr_err_ratelimited("kvm: no hardware support\n");
8382 if (ops->disabled_by_bios()) {
8383 pr_err_ratelimited("kvm: disabled by bios\n");
8389 * KVM explicitly assumes that the guest has an FPU and
8390 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8391 * vCPU's FPU state as a fxregs_state struct.
8393 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8394 printk(KERN_ERR "kvm: inadequate fpu\n");
8400 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
8401 __alignof__(struct fpu), SLAB_ACCOUNT,
8403 if (!x86_fpu_cache) {
8404 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
8408 x86_emulator_cache = kvm_alloc_emulator_cache();
8409 if (!x86_emulator_cache) {
8410 pr_err("kvm: failed to allocate cache for x86 emulator\n");
8411 goto out_free_x86_fpu_cache;
8414 user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8415 if (!user_return_msrs) {
8416 printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8417 goto out_free_x86_emulator_cache;
8419 kvm_nr_uret_msrs = 0;
8421 r = kvm_mmu_module_init();
8423 goto out_free_percpu;
8427 perf_register_guest_info_callbacks(&kvm_guest_cbs);
8429 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8430 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8431 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8434 if (pi_inject_timer == -1)
8435 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
8436 #ifdef CONFIG_X86_64
8437 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8439 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8440 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8446 free_percpu(user_return_msrs);
8447 out_free_x86_emulator_cache:
8448 kmem_cache_destroy(x86_emulator_cache);
8449 out_free_x86_fpu_cache:
8450 kmem_cache_destroy(x86_fpu_cache);
8455 void kvm_arch_exit(void)
8457 #ifdef CONFIG_X86_64
8458 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8459 clear_hv_tscchange_cb();
8462 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
8464 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8465 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8466 CPUFREQ_TRANSITION_NOTIFIER);
8467 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8468 #ifdef CONFIG_X86_64
8469 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8470 irq_work_sync(&pvclock_irq_work);
8471 cancel_work_sync(&pvclock_gtod_work);
8473 kvm_x86_ops.hardware_enable = NULL;
8474 kvm_mmu_module_exit();
8475 free_percpu(user_return_msrs);
8476 kmem_cache_destroy(x86_emulator_cache);
8477 kmem_cache_destroy(x86_fpu_cache);
8478 #ifdef CONFIG_KVM_XEN
8479 static_key_deferred_flush(&kvm_xen_enabled);
8480 WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8484 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
8486 ++vcpu->stat.halt_exits;
8487 if (lapic_in_kernel(vcpu)) {
8488 vcpu->arch.mp_state = state;
8491 vcpu->run->exit_reason = reason;
8496 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8498 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8500 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8502 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8504 int ret = kvm_skip_emulated_instruction(vcpu);
8506 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8507 * KVM_EXIT_DEBUG here.
8509 return kvm_vcpu_halt(vcpu) && ret;
8511 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8513 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8515 int ret = kvm_skip_emulated_instruction(vcpu);
8517 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8519 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8521 #ifdef CONFIG_X86_64
8522 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8523 unsigned long clock_type)
8525 struct kvm_clock_pairing clock_pairing;
8526 struct timespec64 ts;
8530 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8531 return -KVM_EOPNOTSUPP;
8533 if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8534 return -KVM_EOPNOTSUPP;
8536 clock_pairing.sec = ts.tv_sec;
8537 clock_pairing.nsec = ts.tv_nsec;
8538 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8539 clock_pairing.flags = 0;
8540 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8543 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8544 sizeof(struct kvm_clock_pairing)))
8552 * kvm_pv_kick_cpu_op: Kick a vcpu.
8554 * @apicid - apicid of vcpu to be kicked.
8556 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8558 struct kvm_lapic_irq lapic_irq;
8560 lapic_irq.shorthand = APIC_DEST_NOSHORT;
8561 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8562 lapic_irq.level = 0;
8563 lapic_irq.dest_id = apicid;
8564 lapic_irq.msi_redir_hint = false;
8566 lapic_irq.delivery_mode = APIC_DM_REMRD;
8567 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8570 bool kvm_apicv_activated(struct kvm *kvm)
8572 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8574 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8576 static void kvm_apicv_init(struct kvm *kvm)
8578 mutex_init(&kvm->arch.apicv_update_lock);
8581 clear_bit(APICV_INHIBIT_REASON_DISABLE,
8582 &kvm->arch.apicv_inhibit_reasons);
8584 set_bit(APICV_INHIBIT_REASON_DISABLE,
8585 &kvm->arch.apicv_inhibit_reasons);
8588 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
8590 struct kvm_vcpu *target = NULL;
8591 struct kvm_apic_map *map;
8593 vcpu->stat.directed_yield_attempted++;
8595 if (single_task_running())
8599 map = rcu_dereference(vcpu->kvm->arch.apic_map);
8601 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8602 target = map->phys_map[dest_id]->vcpu;
8606 if (!target || !READ_ONCE(target->ready))
8609 /* Ignore requests to yield to self */
8613 if (kvm_vcpu_yield_to(target) <= 0)
8616 vcpu->stat.directed_yield_successful++;
8622 static int complete_hypercall_exit(struct kvm_vcpu *vcpu)
8624 u64 ret = vcpu->run->hypercall.ret;
8626 if (!is_64_bit_mode(vcpu))
8628 kvm_rax_write(vcpu, ret);
8629 ++vcpu->stat.hypercalls;
8630 return kvm_skip_emulated_instruction(vcpu);
8633 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8635 unsigned long nr, a0, a1, a2, a3, ret;
8638 if (kvm_xen_hypercall_enabled(vcpu->kvm))
8639 return kvm_xen_hypercall(vcpu);
8641 if (kvm_hv_hypercall_enabled(vcpu))
8642 return kvm_hv_hypercall(vcpu);
8644 nr = kvm_rax_read(vcpu);
8645 a0 = kvm_rbx_read(vcpu);
8646 a1 = kvm_rcx_read(vcpu);
8647 a2 = kvm_rdx_read(vcpu);
8648 a3 = kvm_rsi_read(vcpu);
8650 trace_kvm_hypercall(nr, a0, a1, a2, a3);
8652 op_64_bit = is_64_bit_mode(vcpu);
8661 if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
8669 case KVM_HC_VAPIC_POLL_IRQ:
8672 case KVM_HC_KICK_CPU:
8673 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8676 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8677 kvm_sched_yield(vcpu, a1);
8680 #ifdef CONFIG_X86_64
8681 case KVM_HC_CLOCK_PAIRING:
8682 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8685 case KVM_HC_SEND_IPI:
8686 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8689 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8691 case KVM_HC_SCHED_YIELD:
8692 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8695 kvm_sched_yield(vcpu, a0);
8698 case KVM_HC_MAP_GPA_RANGE: {
8699 u64 gpa = a0, npages = a1, attrs = a2;
8702 if (!(vcpu->kvm->arch.hypercall_exit_enabled & (1 << KVM_HC_MAP_GPA_RANGE)))
8705 if (!PAGE_ALIGNED(gpa) || !npages ||
8706 gpa_to_gfn(gpa) + npages <= gpa_to_gfn(gpa)) {
8711 vcpu->run->exit_reason = KVM_EXIT_HYPERCALL;
8712 vcpu->run->hypercall.nr = KVM_HC_MAP_GPA_RANGE;
8713 vcpu->run->hypercall.args[0] = gpa;
8714 vcpu->run->hypercall.args[1] = npages;
8715 vcpu->run->hypercall.args[2] = attrs;
8716 vcpu->run->hypercall.longmode = op_64_bit;
8717 vcpu->arch.complete_userspace_io = complete_hypercall_exit;
8727 kvm_rax_write(vcpu, ret);
8729 ++vcpu->stat.hypercalls;
8730 return kvm_skip_emulated_instruction(vcpu);
8732 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8734 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8736 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8737 char instruction[3];
8738 unsigned long rip = kvm_rip_read(vcpu);
8740 static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
8742 return emulator_write_emulated(ctxt, rip, instruction, 3,
8746 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8748 return vcpu->run->request_interrupt_window &&
8749 likely(!pic_in_kernel(vcpu->kvm));
8752 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8754 struct kvm_run *kvm_run = vcpu->run;
8757 * if_flag is obsolete and useless, so do not bother
8758 * setting it for SEV-ES guests. Userspace can just
8759 * use kvm_run->ready_for_interrupt_injection.
8761 kvm_run->if_flag = !vcpu->arch.guest_state_protected
8762 && (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8764 kvm_run->cr8 = kvm_get_cr8(vcpu);
8765 kvm_run->apic_base = kvm_get_apic_base(vcpu);
8766 kvm_run->ready_for_interrupt_injection =
8767 pic_in_kernel(vcpu->kvm) ||
8768 kvm_vcpu_ready_for_interrupt_injection(vcpu);
8771 kvm_run->flags |= KVM_RUN_X86_SMM;
8774 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8778 if (!kvm_x86_ops.update_cr8_intercept)
8781 if (!lapic_in_kernel(vcpu))
8784 if (vcpu->arch.apicv_active)
8787 if (!vcpu->arch.apic->vapic_addr)
8788 max_irr = kvm_lapic_find_highest_irr(vcpu);
8795 tpr = kvm_lapic_get_cr8(vcpu);
8797 static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
8801 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
8803 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8804 kvm_x86_ops.nested_ops->triple_fault(vcpu);
8808 return kvm_x86_ops.nested_ops->check_events(vcpu);
8811 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
8813 if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
8814 vcpu->arch.exception.error_code = false;
8815 static_call(kvm_x86_queue_exception)(vcpu);
8818 static int inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8821 bool can_inject = true;
8823 /* try to reinject previous events if any */
8825 if (vcpu->arch.exception.injected) {
8826 kvm_inject_exception(vcpu);
8830 * Do not inject an NMI or interrupt if there is a pending
8831 * exception. Exceptions and interrupts are recognized at
8832 * instruction boundaries, i.e. the start of an instruction.
8833 * Trap-like exceptions, e.g. #DB, have higher priority than
8834 * NMIs and interrupts, i.e. traps are recognized before an
8835 * NMI/interrupt that's pending on the same instruction.
8836 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8837 * priority, but are only generated (pended) during instruction
8838 * execution, i.e. a pending fault-like exception means the
8839 * fault occurred on the *previous* instruction and must be
8840 * serviced prior to recognizing any new events in order to
8841 * fully complete the previous instruction.
8843 else if (!vcpu->arch.exception.pending) {
8844 if (vcpu->arch.nmi_injected) {
8845 static_call(kvm_x86_set_nmi)(vcpu);
8847 } else if (vcpu->arch.interrupt.injected) {
8848 static_call(kvm_x86_set_irq)(vcpu);
8853 WARN_ON_ONCE(vcpu->arch.exception.injected &&
8854 vcpu->arch.exception.pending);
8857 * Call check_nested_events() even if we reinjected a previous event
8858 * in order for caller to determine if it should require immediate-exit
8859 * from L2 to L1 due to pending L1 events which require exit
8862 if (is_guest_mode(vcpu)) {
8863 r = kvm_check_nested_events(vcpu);
8868 /* try to inject new event if pending */
8869 if (vcpu->arch.exception.pending) {
8870 trace_kvm_inj_exception(vcpu->arch.exception.nr,
8871 vcpu->arch.exception.has_error_code,
8872 vcpu->arch.exception.error_code);
8874 vcpu->arch.exception.pending = false;
8875 vcpu->arch.exception.injected = true;
8877 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
8878 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
8881 if (vcpu->arch.exception.nr == DB_VECTOR) {
8882 kvm_deliver_exception_payload(vcpu);
8883 if (vcpu->arch.dr7 & DR7_GD) {
8884 vcpu->arch.dr7 &= ~DR7_GD;
8885 kvm_update_dr7(vcpu);
8889 kvm_inject_exception(vcpu);
8893 /* Don't inject interrupts if the user asked to avoid doing so */
8894 if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ)
8898 * Finally, inject interrupt events. If an event cannot be injected
8899 * due to architectural conditions (e.g. IF=0) a window-open exit
8900 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
8901 * and can architecturally be injected, but we cannot do it right now:
8902 * an interrupt could have arrived just now and we have to inject it
8903 * as a vmexit, or there could already an event in the queue, which is
8904 * indicated by can_inject. In that case we request an immediate exit
8905 * in order to make progress and get back here for another iteration.
8906 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8908 if (vcpu->arch.smi_pending) {
8909 r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
8913 vcpu->arch.smi_pending = false;
8914 ++vcpu->arch.smi_count;
8918 static_call(kvm_x86_enable_smi_window)(vcpu);
8921 if (vcpu->arch.nmi_pending) {
8922 r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
8926 --vcpu->arch.nmi_pending;
8927 vcpu->arch.nmi_injected = true;
8928 static_call(kvm_x86_set_nmi)(vcpu);
8930 WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
8932 if (vcpu->arch.nmi_pending)
8933 static_call(kvm_x86_enable_nmi_window)(vcpu);
8936 if (kvm_cpu_has_injectable_intr(vcpu)) {
8937 r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
8941 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
8942 static_call(kvm_x86_set_irq)(vcpu);
8943 WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
8945 if (kvm_cpu_has_injectable_intr(vcpu))
8946 static_call(kvm_x86_enable_irq_window)(vcpu);
8949 if (is_guest_mode(vcpu) &&
8950 kvm_x86_ops.nested_ops->hv_timer_pending &&
8951 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
8952 *req_immediate_exit = true;
8954 WARN_ON(vcpu->arch.exception.pending);
8959 *req_immediate_exit = true;
8965 static void process_nmi(struct kvm_vcpu *vcpu)
8970 * x86 is limited to one NMI running, and one NMI pending after it.
8971 * If an NMI is already in progress, limit further NMIs to just one.
8972 * Otherwise, allow two (and we'll inject the first one immediately).
8974 if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
8977 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
8978 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
8979 kvm_make_request(KVM_REQ_EVENT, vcpu);
8982 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
8985 flags |= seg->g << 23;
8986 flags |= seg->db << 22;
8987 flags |= seg->l << 21;
8988 flags |= seg->avl << 20;
8989 flags |= seg->present << 15;
8990 flags |= seg->dpl << 13;
8991 flags |= seg->s << 12;
8992 flags |= seg->type << 8;
8996 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
8998 struct kvm_segment seg;
9001 kvm_get_segment(vcpu, &seg, n);
9002 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
9005 offset = 0x7f84 + n * 12;
9007 offset = 0x7f2c + (n - 3) * 12;
9009 put_smstate(u32, buf, offset + 8, seg.base);
9010 put_smstate(u32, buf, offset + 4, seg.limit);
9011 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
9014 #ifdef CONFIG_X86_64
9015 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
9017 struct kvm_segment seg;
9021 kvm_get_segment(vcpu, &seg, n);
9022 offset = 0x7e00 + n * 16;
9024 flags = enter_smm_get_segment_flags(&seg) >> 8;
9025 put_smstate(u16, buf, offset, seg.selector);
9026 put_smstate(u16, buf, offset + 2, flags);
9027 put_smstate(u32, buf, offset + 4, seg.limit);
9028 put_smstate(u64, buf, offset + 8, seg.base);
9032 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
9035 struct kvm_segment seg;
9039 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
9040 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
9041 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
9042 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
9044 for (i = 0; i < 8; i++)
9045 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i));
9047 kvm_get_dr(vcpu, 6, &val);
9048 put_smstate(u32, buf, 0x7fcc, (u32)val);
9049 kvm_get_dr(vcpu, 7, &val);
9050 put_smstate(u32, buf, 0x7fc8, (u32)val);
9052 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9053 put_smstate(u32, buf, 0x7fc4, seg.selector);
9054 put_smstate(u32, buf, 0x7f64, seg.base);
9055 put_smstate(u32, buf, 0x7f60, seg.limit);
9056 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
9058 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9059 put_smstate(u32, buf, 0x7fc0, seg.selector);
9060 put_smstate(u32, buf, 0x7f80, seg.base);
9061 put_smstate(u32, buf, 0x7f7c, seg.limit);
9062 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
9064 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9065 put_smstate(u32, buf, 0x7f74, dt.address);
9066 put_smstate(u32, buf, 0x7f70, dt.size);
9068 static_call(kvm_x86_get_idt)(vcpu, &dt);
9069 put_smstate(u32, buf, 0x7f58, dt.address);
9070 put_smstate(u32, buf, 0x7f54, dt.size);
9072 for (i = 0; i < 6; i++)
9073 enter_smm_save_seg_32(vcpu, buf, i);
9075 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
9078 put_smstate(u32, buf, 0x7efc, 0x00020000);
9079 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
9082 #ifdef CONFIG_X86_64
9083 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
9086 struct kvm_segment seg;
9090 for (i = 0; i < 16; i++)
9091 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i));
9093 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
9094 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
9096 kvm_get_dr(vcpu, 6, &val);
9097 put_smstate(u64, buf, 0x7f68, val);
9098 kvm_get_dr(vcpu, 7, &val);
9099 put_smstate(u64, buf, 0x7f60, val);
9101 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
9102 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
9103 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
9105 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
9108 put_smstate(u32, buf, 0x7efc, 0x00020064);
9110 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
9112 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9113 put_smstate(u16, buf, 0x7e90, seg.selector);
9114 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
9115 put_smstate(u32, buf, 0x7e94, seg.limit);
9116 put_smstate(u64, buf, 0x7e98, seg.base);
9118 static_call(kvm_x86_get_idt)(vcpu, &dt);
9119 put_smstate(u32, buf, 0x7e84, dt.size);
9120 put_smstate(u64, buf, 0x7e88, dt.address);
9122 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9123 put_smstate(u16, buf, 0x7e70, seg.selector);
9124 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
9125 put_smstate(u32, buf, 0x7e74, seg.limit);
9126 put_smstate(u64, buf, 0x7e78, seg.base);
9128 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9129 put_smstate(u32, buf, 0x7e64, dt.size);
9130 put_smstate(u64, buf, 0x7e68, dt.address);
9132 for (i = 0; i < 6; i++)
9133 enter_smm_save_seg_64(vcpu, buf, i);
9137 static void enter_smm(struct kvm_vcpu *vcpu)
9139 struct kvm_segment cs, ds;
9144 memset(buf, 0, 512);
9145 #ifdef CONFIG_X86_64
9146 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9147 enter_smm_save_state_64(vcpu, buf);
9150 enter_smm_save_state_32(vcpu, buf);
9153 * Give enter_smm() a chance to make ISA-specific changes to the vCPU
9154 * state (e.g. leave guest mode) after we've saved the state into the
9155 * SMM state-save area.
9157 static_call(kvm_x86_enter_smm)(vcpu, buf);
9159 kvm_smm_changed(vcpu, true);
9160 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
9162 if (static_call(kvm_x86_get_nmi_mask)(vcpu))
9163 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
9165 static_call(kvm_x86_set_nmi_mask)(vcpu, true);
9167 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
9168 kvm_rip_write(vcpu, 0x8000);
9170 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
9171 static_call(kvm_x86_set_cr0)(vcpu, cr0);
9172 vcpu->arch.cr0 = cr0;
9174 static_call(kvm_x86_set_cr4)(vcpu, 0);
9176 /* Undocumented: IDT limit is set to zero on entry to SMM. */
9177 dt.address = dt.size = 0;
9178 static_call(kvm_x86_set_idt)(vcpu, &dt);
9180 kvm_set_dr(vcpu, 7, DR7_FIXED_1);
9182 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
9183 cs.base = vcpu->arch.smbase;
9188 cs.limit = ds.limit = 0xffffffff;
9189 cs.type = ds.type = 0x3;
9190 cs.dpl = ds.dpl = 0;
9195 cs.avl = ds.avl = 0;
9196 cs.present = ds.present = 1;
9197 cs.unusable = ds.unusable = 0;
9198 cs.padding = ds.padding = 0;
9200 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9201 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
9202 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
9203 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
9204 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
9205 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
9207 #ifdef CONFIG_X86_64
9208 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9209 static_call(kvm_x86_set_efer)(vcpu, 0);
9212 kvm_update_cpuid_runtime(vcpu);
9213 kvm_mmu_reset_context(vcpu);
9216 static void process_smi(struct kvm_vcpu *vcpu)
9218 vcpu->arch.smi_pending = true;
9219 kvm_make_request(KVM_REQ_EVENT, vcpu);
9222 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
9223 unsigned long *vcpu_bitmap)
9225 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC, vcpu_bitmap);
9228 void kvm_make_scan_ioapic_request(struct kvm *kvm)
9230 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
9233 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
9237 if (!lapic_in_kernel(vcpu))
9240 mutex_lock(&vcpu->kvm->arch.apicv_update_lock);
9242 activate = kvm_apicv_activated(vcpu->kvm);
9243 if (vcpu->arch.apicv_active == activate)
9246 vcpu->arch.apicv_active = activate;
9247 kvm_apic_update_apicv(vcpu);
9248 static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
9251 * When APICv gets disabled, we may still have injected interrupts
9252 * pending. At the same time, KVM_REQ_EVENT may not be set as APICv was
9253 * still active when the interrupt got accepted. Make sure
9254 * inject_pending_event() is called to check for that.
9256 if (!vcpu->arch.apicv_active)
9257 kvm_make_request(KVM_REQ_EVENT, vcpu);
9260 mutex_unlock(&vcpu->kvm->arch.apicv_update_lock);
9262 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
9264 void __kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9266 unsigned long old, new;
9268 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
9269 !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
9272 old = new = kvm->arch.apicv_inhibit_reasons;
9275 __clear_bit(bit, &new);
9277 __set_bit(bit, &new);
9279 if (!!old != !!new) {
9280 trace_kvm_apicv_update_request(activate, bit);
9281 kvm_make_all_cpus_request(kvm, KVM_REQ_APICV_UPDATE);
9282 kvm->arch.apicv_inhibit_reasons = new;
9284 unsigned long gfn = gpa_to_gfn(APIC_DEFAULT_PHYS_BASE);
9285 kvm_zap_gfn_range(kvm, gfn, gfn+1);
9288 kvm->arch.apicv_inhibit_reasons = new;
9290 EXPORT_SYMBOL_GPL(__kvm_request_apicv_update);
9292 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9294 mutex_lock(&kvm->arch.apicv_update_lock);
9295 __kvm_request_apicv_update(kvm, activate, bit);
9296 mutex_unlock(&kvm->arch.apicv_update_lock);
9298 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
9300 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
9302 if (!kvm_apic_present(vcpu))
9305 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
9307 if (irqchip_split(vcpu->kvm))
9308 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
9310 if (vcpu->arch.apicv_active)
9311 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9312 if (ioapic_in_kernel(vcpu->kvm))
9313 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
9316 if (is_guest_mode(vcpu))
9317 vcpu->arch.load_eoi_exitmap_pending = true;
9319 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
9322 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
9324 u64 eoi_exit_bitmap[4];
9326 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
9329 if (to_hv_vcpu(vcpu))
9330 bitmap_or((ulong *)eoi_exit_bitmap,
9331 vcpu->arch.ioapic_handled_vectors,
9332 to_hv_synic(vcpu)->vec_bitmap, 256);
9334 static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
9337 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
9338 unsigned long start, unsigned long end)
9340 unsigned long apic_address;
9343 * The physical address of apic access page is stored in the VMCS.
9344 * Update it when it becomes invalid.
9346 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
9347 if (start <= apic_address && apic_address < end)
9348 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
9351 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
9353 if (!lapic_in_kernel(vcpu))
9356 if (!kvm_x86_ops.set_apic_access_page_addr)
9359 static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
9362 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
9364 smp_send_reschedule(vcpu->cpu);
9366 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
9369 * Returns 1 to let vcpu_run() continue the guest execution loop without
9370 * exiting to the userspace. Otherwise, the value will be returned to the
9373 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
9377 dm_request_for_irq_injection(vcpu) &&
9378 kvm_cpu_accept_dm_intr(vcpu);
9379 fastpath_t exit_fastpath;
9381 bool req_immediate_exit = false;
9383 /* Forbid vmenter if vcpu dirty ring is soft-full */
9384 if (unlikely(vcpu->kvm->dirty_ring_size &&
9385 kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
9386 vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
9387 trace_kvm_dirty_ring_exit(vcpu);
9392 if (kvm_request_pending(vcpu)) {
9393 if (kvm_check_request(KVM_REQ_VM_BUGGED, vcpu)) {
9397 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
9398 if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
9403 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
9404 kvm_mmu_unload(vcpu);
9405 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
9406 __kvm_migrate_timers(vcpu);
9407 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
9408 kvm_gen_update_masterclock(vcpu->kvm);
9409 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
9410 kvm_gen_kvmclock_update(vcpu);
9411 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
9412 r = kvm_guest_time_update(vcpu);
9416 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
9417 kvm_mmu_sync_roots(vcpu);
9418 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
9419 kvm_mmu_load_pgd(vcpu);
9420 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
9421 kvm_vcpu_flush_tlb_all(vcpu);
9423 /* Flushing all ASIDs flushes the current ASID... */
9424 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
9426 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
9427 kvm_vcpu_flush_tlb_current(vcpu);
9428 if (kvm_check_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu))
9429 kvm_vcpu_flush_tlb_guest(vcpu);
9431 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
9432 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
9436 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9437 if (is_guest_mode(vcpu)) {
9438 kvm_x86_ops.nested_ops->triple_fault(vcpu);
9440 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
9441 vcpu->mmio_needed = 0;
9446 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
9447 /* Page is swapped out. Do synthetic halt */
9448 vcpu->arch.apf.halted = true;
9452 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
9453 record_steal_time(vcpu);
9454 if (kvm_check_request(KVM_REQ_SMI, vcpu))
9456 if (kvm_check_request(KVM_REQ_NMI, vcpu))
9458 if (kvm_check_request(KVM_REQ_PMU, vcpu))
9459 kvm_pmu_handle_event(vcpu);
9460 if (kvm_check_request(KVM_REQ_PMI, vcpu))
9461 kvm_pmu_deliver_pmi(vcpu);
9462 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
9463 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
9464 if (test_bit(vcpu->arch.pending_ioapic_eoi,
9465 vcpu->arch.ioapic_handled_vectors)) {
9466 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
9467 vcpu->run->eoi.vector =
9468 vcpu->arch.pending_ioapic_eoi;
9473 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
9474 vcpu_scan_ioapic(vcpu);
9475 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
9476 vcpu_load_eoi_exitmap(vcpu);
9477 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
9478 kvm_vcpu_reload_apic_access_page(vcpu);
9479 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
9480 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9481 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
9485 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
9486 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9487 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
9491 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
9492 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
9494 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
9495 vcpu->run->hyperv = hv_vcpu->exit;
9501 * KVM_REQ_HV_STIMER has to be processed after
9502 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
9503 * depend on the guest clock being up-to-date
9505 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
9506 kvm_hv_process_stimers(vcpu);
9507 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
9508 kvm_vcpu_update_apicv(vcpu);
9509 if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
9510 kvm_check_async_pf_completion(vcpu);
9511 if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
9512 static_call(kvm_x86_msr_filter_changed)(vcpu);
9514 if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
9515 static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
9518 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
9519 kvm_xen_has_interrupt(vcpu)) {
9520 ++vcpu->stat.req_event;
9521 r = kvm_apic_accept_events(vcpu);
9526 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
9531 r = inject_pending_event(vcpu, &req_immediate_exit);
9537 static_call(kvm_x86_enable_irq_window)(vcpu);
9539 if (kvm_lapic_enabled(vcpu)) {
9540 update_cr8_intercept(vcpu);
9541 kvm_lapic_sync_to_vapic(vcpu);
9545 r = kvm_mmu_reload(vcpu);
9547 goto cancel_injection;
9552 static_call(kvm_x86_prepare_guest_switch)(vcpu);
9555 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
9556 * IPI are then delayed after guest entry, which ensures that they
9557 * result in virtual interrupt delivery.
9559 local_irq_disable();
9560 vcpu->mode = IN_GUEST_MODE;
9562 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9565 * 1) We should set ->mode before checking ->requests. Please see
9566 * the comment in kvm_vcpu_exiting_guest_mode().
9568 * 2) For APICv, we should set ->mode before checking PID.ON. This
9569 * pairs with the memory barrier implicit in pi_test_and_set_on
9570 * (see vmx_deliver_posted_interrupt).
9572 * 3) This also orders the write to mode from any reads to the page
9573 * tables done while the VCPU is running. Please see the comment
9574 * in kvm_flush_remote_tlbs.
9576 smp_mb__after_srcu_read_unlock();
9579 * This handles the case where a posted interrupt was
9580 * notified with kvm_vcpu_kick.
9582 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
9583 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9585 if (kvm_vcpu_exit_request(vcpu)) {
9586 vcpu->mode = OUTSIDE_GUEST_MODE;
9590 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9592 goto cancel_injection;
9595 if (req_immediate_exit) {
9596 kvm_make_request(KVM_REQ_EVENT, vcpu);
9597 static_call(kvm_x86_request_immediate_exit)(vcpu);
9600 fpregs_assert_state_consistent();
9601 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9602 switch_fpu_return();
9604 if (unlikely(vcpu->arch.switch_db_regs)) {
9606 set_debugreg(vcpu->arch.eff_db[0], 0);
9607 set_debugreg(vcpu->arch.eff_db[1], 1);
9608 set_debugreg(vcpu->arch.eff_db[2], 2);
9609 set_debugreg(vcpu->arch.eff_db[3], 3);
9610 } else if (unlikely(hw_breakpoint_active())) {
9615 exit_fastpath = static_call(kvm_x86_run)(vcpu);
9616 if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9619 if (unlikely(kvm_vcpu_exit_request(vcpu))) {
9620 exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
9624 if (vcpu->arch.apicv_active)
9625 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9629 * Do this here before restoring debug registers on the host. And
9630 * since we do this before handling the vmexit, a DR access vmexit
9631 * can (a) read the correct value of the debug registers, (b) set
9632 * KVM_DEBUGREG_WONT_EXIT again.
9634 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9635 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9636 static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
9637 kvm_update_dr0123(vcpu);
9638 kvm_update_dr7(vcpu);
9642 * If the guest has used debug registers, at least dr7
9643 * will be disabled while returning to the host.
9644 * If we don't have active breakpoints in the host, we don't
9645 * care about the messed up debug address registers. But if
9646 * we have some of them active, restore the old state.
9648 if (hw_breakpoint_active())
9649 hw_breakpoint_restore();
9651 vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9652 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9654 vcpu->mode = OUTSIDE_GUEST_MODE;
9657 static_call(kvm_x86_handle_exit_irqoff)(vcpu);
9660 * Consume any pending interrupts, including the possible source of
9661 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9662 * An instruction is required after local_irq_enable() to fully unblock
9663 * interrupts on processors that implement an interrupt shadow, the
9664 * stat.exits increment will do nicely.
9666 kvm_before_interrupt(vcpu);
9669 local_irq_disable();
9670 kvm_after_interrupt(vcpu);
9673 * Wait until after servicing IRQs to account guest time so that any
9674 * ticks that occurred while running the guest are properly accounted
9675 * to the guest. Waiting until IRQs are enabled degrades the accuracy
9676 * of accounting via context tracking, but the loss of accuracy is
9677 * acceptable for all known use cases.
9679 vtime_account_guest_exit();
9681 if (lapic_in_kernel(vcpu)) {
9682 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9683 if (delta != S64_MIN) {
9684 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9685 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9692 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9695 * Profile KVM exit RIPs:
9697 if (unlikely(prof_on == KVM_PROFILING)) {
9698 unsigned long rip = kvm_rip_read(vcpu);
9699 profile_hit(KVM_PROFILING, (void *)rip);
9702 if (unlikely(vcpu->arch.tsc_always_catchup))
9703 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9705 if (vcpu->arch.apic_attention)
9706 kvm_lapic_sync_from_vapic(vcpu);
9708 r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
9712 if (req_immediate_exit)
9713 kvm_make_request(KVM_REQ_EVENT, vcpu);
9714 static_call(kvm_x86_cancel_injection)(vcpu);
9715 if (unlikely(vcpu->arch.apic_attention))
9716 kvm_lapic_sync_from_vapic(vcpu);
9721 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9723 if (!kvm_arch_vcpu_runnable(vcpu) &&
9724 (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
9725 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9726 kvm_vcpu_block(vcpu);
9727 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9729 if (kvm_x86_ops.post_block)
9730 static_call(kvm_x86_post_block)(vcpu);
9732 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9736 if (kvm_apic_accept_events(vcpu) < 0)
9738 switch(vcpu->arch.mp_state) {
9739 case KVM_MP_STATE_HALTED:
9740 case KVM_MP_STATE_AP_RESET_HOLD:
9741 vcpu->arch.pv.pv_unhalted = false;
9742 vcpu->arch.mp_state =
9743 KVM_MP_STATE_RUNNABLE;
9745 case KVM_MP_STATE_RUNNABLE:
9746 vcpu->arch.apf.halted = false;
9748 case KVM_MP_STATE_INIT_RECEIVED:
9756 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9758 if (is_guest_mode(vcpu))
9759 kvm_check_nested_events(vcpu);
9761 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9762 !vcpu->arch.apf.halted);
9765 static int vcpu_run(struct kvm_vcpu *vcpu)
9768 struct kvm *kvm = vcpu->kvm;
9770 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9771 vcpu->arch.l1tf_flush_l1d = true;
9774 if (kvm_vcpu_running(vcpu)) {
9775 r = vcpu_enter_guest(vcpu);
9777 r = vcpu_block(kvm, vcpu);
9783 kvm_clear_request(KVM_REQ_UNBLOCK, vcpu);
9784 if (kvm_cpu_has_pending_timer(vcpu))
9785 kvm_inject_pending_timer_irqs(vcpu);
9787 if (dm_request_for_irq_injection(vcpu) &&
9788 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9790 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9791 ++vcpu->stat.request_irq_exits;
9795 if (__xfer_to_guest_mode_work_pending()) {
9796 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9797 r = xfer_to_guest_mode_handle_work(vcpu);
9800 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9804 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9809 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9813 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9814 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9815 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9819 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9821 BUG_ON(!vcpu->arch.pio.count);
9823 return complete_emulated_io(vcpu);
9827 * Implements the following, as a state machine:
9831 * for each mmio piece in the fragment
9839 * for each mmio piece in the fragment
9844 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
9846 struct kvm_run *run = vcpu->run;
9847 struct kvm_mmio_fragment *frag;
9850 BUG_ON(!vcpu->mmio_needed);
9852 /* Complete previous fragment */
9853 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
9854 len = min(8u, frag->len);
9855 if (!vcpu->mmio_is_write)
9856 memcpy(frag->data, run->mmio.data, len);
9858 if (frag->len <= 8) {
9859 /* Switch to the next fragment. */
9861 vcpu->mmio_cur_fragment++;
9863 /* Go forward to the next mmio piece. */
9869 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
9870 vcpu->mmio_needed = 0;
9872 /* FIXME: return into emulator if single-stepping. */
9873 if (vcpu->mmio_is_write)
9875 vcpu->mmio_read_completed = 1;
9876 return complete_emulated_io(vcpu);
9879 run->exit_reason = KVM_EXIT_MMIO;
9880 run->mmio.phys_addr = frag->gpa;
9881 if (vcpu->mmio_is_write)
9882 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
9883 run->mmio.len = min(8u, frag->len);
9884 run->mmio.is_write = vcpu->mmio_is_write;
9885 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9889 static void kvm_save_current_fpu(struct fpu *fpu)
9892 * If the target FPU state is not resident in the CPU registers, just
9893 * memcpy() from current, else save CPU state directly to the target.
9895 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9896 memcpy(&fpu->state, ¤t->thread.fpu.state,
9897 fpu_kernel_xstate_size);
9899 save_fpregs_to_fpstate(fpu);
9902 /* Swap (qemu) user FPU context for the guest FPU context. */
9903 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
9907 kvm_save_current_fpu(vcpu->arch.user_fpu);
9910 * Guests with protected state can't have it set by the hypervisor,
9911 * so skip trying to set it.
9913 if (vcpu->arch.guest_fpu)
9914 /* PKRU is separately restored in kvm_x86_ops.run. */
9915 __restore_fpregs_from_fpstate(&vcpu->arch.guest_fpu->state,
9916 ~XFEATURE_MASK_PKRU);
9918 fpregs_mark_activate();
9924 /* When vcpu_run ends, restore user space FPU context. */
9925 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
9930 * Guests with protected state can't have it read by the hypervisor,
9931 * so skip trying to save it.
9933 if (vcpu->arch.guest_fpu)
9934 kvm_save_current_fpu(vcpu->arch.guest_fpu);
9936 restore_fpregs_from_fpstate(&vcpu->arch.user_fpu->state);
9938 fpregs_mark_activate();
9941 ++vcpu->stat.fpu_reload;
9945 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
9947 struct kvm_run *kvm_run = vcpu->run;
9951 kvm_sigset_activate(vcpu);
9953 kvm_load_guest_fpu(vcpu);
9955 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
9956 if (kvm_run->immediate_exit) {
9960 kvm_vcpu_block(vcpu);
9961 if (kvm_apic_accept_events(vcpu) < 0) {
9965 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
9967 if (signal_pending(current)) {
9969 kvm_run->exit_reason = KVM_EXIT_INTR;
9970 ++vcpu->stat.signal_exits;
9975 if ((kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) ||
9976 (kvm_run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)) {
9981 if (kvm_run->kvm_dirty_regs) {
9982 r = sync_regs(vcpu);
9987 /* re-sync apic's tpr */
9988 if (!lapic_in_kernel(vcpu)) {
9989 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
9995 if (unlikely(vcpu->arch.complete_userspace_io)) {
9996 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
9997 vcpu->arch.complete_userspace_io = NULL;
10002 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
10004 if (kvm_run->immediate_exit)
10007 r = vcpu_run(vcpu);
10010 kvm_put_guest_fpu(vcpu);
10011 if (kvm_run->kvm_valid_regs)
10013 post_kvm_run_save(vcpu);
10014 kvm_sigset_deactivate(vcpu);
10020 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10022 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
10024 * We are here if userspace calls get_regs() in the middle of
10025 * instruction emulation. Registers state needs to be copied
10026 * back from emulation context to vcpu. Userspace shouldn't do
10027 * that usually, but some bad designed PV devices (vmware
10028 * backdoor interface) need this to work
10030 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
10031 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10033 regs->rax = kvm_rax_read(vcpu);
10034 regs->rbx = kvm_rbx_read(vcpu);
10035 regs->rcx = kvm_rcx_read(vcpu);
10036 regs->rdx = kvm_rdx_read(vcpu);
10037 regs->rsi = kvm_rsi_read(vcpu);
10038 regs->rdi = kvm_rdi_read(vcpu);
10039 regs->rsp = kvm_rsp_read(vcpu);
10040 regs->rbp = kvm_rbp_read(vcpu);
10041 #ifdef CONFIG_X86_64
10042 regs->r8 = kvm_r8_read(vcpu);
10043 regs->r9 = kvm_r9_read(vcpu);
10044 regs->r10 = kvm_r10_read(vcpu);
10045 regs->r11 = kvm_r11_read(vcpu);
10046 regs->r12 = kvm_r12_read(vcpu);
10047 regs->r13 = kvm_r13_read(vcpu);
10048 regs->r14 = kvm_r14_read(vcpu);
10049 regs->r15 = kvm_r15_read(vcpu);
10052 regs->rip = kvm_rip_read(vcpu);
10053 regs->rflags = kvm_get_rflags(vcpu);
10056 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10059 __get_regs(vcpu, regs);
10064 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10066 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
10067 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10069 kvm_rax_write(vcpu, regs->rax);
10070 kvm_rbx_write(vcpu, regs->rbx);
10071 kvm_rcx_write(vcpu, regs->rcx);
10072 kvm_rdx_write(vcpu, regs->rdx);
10073 kvm_rsi_write(vcpu, regs->rsi);
10074 kvm_rdi_write(vcpu, regs->rdi);
10075 kvm_rsp_write(vcpu, regs->rsp);
10076 kvm_rbp_write(vcpu, regs->rbp);
10077 #ifdef CONFIG_X86_64
10078 kvm_r8_write(vcpu, regs->r8);
10079 kvm_r9_write(vcpu, regs->r9);
10080 kvm_r10_write(vcpu, regs->r10);
10081 kvm_r11_write(vcpu, regs->r11);
10082 kvm_r12_write(vcpu, regs->r12);
10083 kvm_r13_write(vcpu, regs->r13);
10084 kvm_r14_write(vcpu, regs->r14);
10085 kvm_r15_write(vcpu, regs->r15);
10088 kvm_rip_write(vcpu, regs->rip);
10089 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
10091 vcpu->arch.exception.pending = false;
10093 kvm_make_request(KVM_REQ_EVENT, vcpu);
10096 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10099 __set_regs(vcpu, regs);
10104 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
10106 struct kvm_segment cs;
10108 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10112 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
10114 static void __get_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10116 struct desc_ptr dt;
10118 if (vcpu->arch.guest_state_protected)
10119 goto skip_protected_regs;
10121 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10122 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10123 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10124 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10125 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10126 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10128 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10129 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10131 static_call(kvm_x86_get_idt)(vcpu, &dt);
10132 sregs->idt.limit = dt.size;
10133 sregs->idt.base = dt.address;
10134 static_call(kvm_x86_get_gdt)(vcpu, &dt);
10135 sregs->gdt.limit = dt.size;
10136 sregs->gdt.base = dt.address;
10138 sregs->cr2 = vcpu->arch.cr2;
10139 sregs->cr3 = kvm_read_cr3(vcpu);
10141 skip_protected_regs:
10142 sregs->cr0 = kvm_read_cr0(vcpu);
10143 sregs->cr4 = kvm_read_cr4(vcpu);
10144 sregs->cr8 = kvm_get_cr8(vcpu);
10145 sregs->efer = vcpu->arch.efer;
10146 sregs->apic_base = kvm_get_apic_base(vcpu);
10149 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10151 __get_sregs_common(vcpu, sregs);
10153 if (vcpu->arch.guest_state_protected)
10156 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
10157 set_bit(vcpu->arch.interrupt.nr,
10158 (unsigned long *)sregs->interrupt_bitmap);
10161 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10165 __get_sregs_common(vcpu, (struct kvm_sregs *)sregs2);
10167 if (vcpu->arch.guest_state_protected)
10170 if (is_pae_paging(vcpu)) {
10171 for (i = 0 ; i < 4 ; i++)
10172 sregs2->pdptrs[i] = kvm_pdptr_read(vcpu, i);
10173 sregs2->flags |= KVM_SREGS2_FLAGS_PDPTRS_VALID;
10177 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
10178 struct kvm_sregs *sregs)
10181 __get_sregs(vcpu, sregs);
10186 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
10187 struct kvm_mp_state *mp_state)
10192 if (kvm_mpx_supported())
10193 kvm_load_guest_fpu(vcpu);
10195 r = kvm_apic_accept_events(vcpu);
10200 if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
10201 vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
10202 vcpu->arch.pv.pv_unhalted)
10203 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
10205 mp_state->mp_state = vcpu->arch.mp_state;
10208 if (kvm_mpx_supported())
10209 kvm_put_guest_fpu(vcpu);
10214 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
10215 struct kvm_mp_state *mp_state)
10221 if (!lapic_in_kernel(vcpu) &&
10222 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
10226 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
10227 * INIT state; latched init should be reported using
10228 * KVM_SET_VCPU_EVENTS, so reject it here.
10230 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
10231 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
10232 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
10235 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
10236 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
10237 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
10239 vcpu->arch.mp_state = mp_state->mp_state;
10240 kvm_make_request(KVM_REQ_EVENT, vcpu);
10248 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
10249 int reason, bool has_error_code, u32 error_code)
10251 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
10254 init_emulate_ctxt(vcpu);
10256 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
10257 has_error_code, error_code);
10259 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
10260 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
10261 vcpu->run->internal.ndata = 0;
10265 kvm_rip_write(vcpu, ctxt->eip);
10266 kvm_set_rflags(vcpu, ctxt->eflags);
10269 EXPORT_SYMBOL_GPL(kvm_task_switch);
10271 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10273 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
10275 * When EFER.LME and CR0.PG are set, the processor is in
10276 * 64-bit mode (though maybe in a 32-bit code segment).
10277 * CR4.PAE and EFER.LMA must be set.
10279 if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
10281 if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
10285 * Not in 64-bit mode: EFER.LMA is clear and the code
10286 * segment cannot be 64-bit.
10288 if (sregs->efer & EFER_LMA || sregs->cs.l)
10292 return kvm_is_valid_cr4(vcpu, sregs->cr4);
10295 static int __set_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs,
10296 int *mmu_reset_needed, bool update_pdptrs)
10298 struct msr_data apic_base_msr;
10300 struct desc_ptr dt;
10302 if (!kvm_is_valid_sregs(vcpu, sregs))
10305 apic_base_msr.data = sregs->apic_base;
10306 apic_base_msr.host_initiated = true;
10307 if (kvm_set_apic_base(vcpu, &apic_base_msr))
10310 if (vcpu->arch.guest_state_protected)
10313 dt.size = sregs->idt.limit;
10314 dt.address = sregs->idt.base;
10315 static_call(kvm_x86_set_idt)(vcpu, &dt);
10316 dt.size = sregs->gdt.limit;
10317 dt.address = sregs->gdt.base;
10318 static_call(kvm_x86_set_gdt)(vcpu, &dt);
10320 vcpu->arch.cr2 = sregs->cr2;
10321 *mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
10322 vcpu->arch.cr3 = sregs->cr3;
10323 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
10325 kvm_set_cr8(vcpu, sregs->cr8);
10327 *mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
10328 static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
10330 *mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
10331 static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
10332 vcpu->arch.cr0 = sregs->cr0;
10334 *mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
10335 static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
10337 if (update_pdptrs) {
10338 idx = srcu_read_lock(&vcpu->kvm->srcu);
10339 if (is_pae_paging(vcpu)) {
10340 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
10341 *mmu_reset_needed = 1;
10343 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10346 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10347 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10348 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10349 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10350 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10351 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10353 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10354 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10356 update_cr8_intercept(vcpu);
10358 /* Older userspace won't unhalt the vcpu on reset. */
10359 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
10360 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
10361 !is_protmode(vcpu))
10362 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10367 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10369 int pending_vec, max_bits;
10370 int mmu_reset_needed = 0;
10371 int ret = __set_sregs_common(vcpu, sregs, &mmu_reset_needed, true);
10376 if (mmu_reset_needed)
10377 kvm_mmu_reset_context(vcpu);
10379 max_bits = KVM_NR_INTERRUPTS;
10380 pending_vec = find_first_bit(
10381 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
10383 if (pending_vec < max_bits) {
10384 kvm_queue_interrupt(vcpu, pending_vec, false);
10385 pr_debug("Set back pending irq %d\n", pending_vec);
10386 kvm_make_request(KVM_REQ_EVENT, vcpu);
10391 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10393 int mmu_reset_needed = 0;
10394 bool valid_pdptrs = sregs2->flags & KVM_SREGS2_FLAGS_PDPTRS_VALID;
10395 bool pae = (sregs2->cr0 & X86_CR0_PG) && (sregs2->cr4 & X86_CR4_PAE) &&
10396 !(sregs2->efer & EFER_LMA);
10399 if (sregs2->flags & ~KVM_SREGS2_FLAGS_PDPTRS_VALID)
10402 if (valid_pdptrs && (!pae || vcpu->arch.guest_state_protected))
10405 ret = __set_sregs_common(vcpu, (struct kvm_sregs *)sregs2,
10406 &mmu_reset_needed, !valid_pdptrs);
10410 if (valid_pdptrs) {
10411 for (i = 0; i < 4 ; i++)
10412 kvm_pdptr_write(vcpu, i, sregs2->pdptrs[i]);
10414 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
10415 mmu_reset_needed = 1;
10416 vcpu->arch.pdptrs_from_userspace = true;
10418 if (mmu_reset_needed)
10419 kvm_mmu_reset_context(vcpu);
10423 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
10424 struct kvm_sregs *sregs)
10429 ret = __set_sregs(vcpu, sregs);
10434 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
10435 struct kvm_guest_debug *dbg)
10437 unsigned long rflags;
10440 if (vcpu->arch.guest_state_protected)
10445 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
10447 if (vcpu->arch.exception.pending)
10449 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
10450 kvm_queue_exception(vcpu, DB_VECTOR);
10452 kvm_queue_exception(vcpu, BP_VECTOR);
10456 * Read rflags as long as potentially injected trace flags are still
10459 rflags = kvm_get_rflags(vcpu);
10461 vcpu->guest_debug = dbg->control;
10462 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
10463 vcpu->guest_debug = 0;
10465 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
10466 for (i = 0; i < KVM_NR_DB_REGS; ++i)
10467 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
10468 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
10470 for (i = 0; i < KVM_NR_DB_REGS; i++)
10471 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
10473 kvm_update_dr7(vcpu);
10475 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10476 vcpu->arch.singlestep_rip = kvm_get_linear_rip(vcpu);
10479 * Trigger an rflags update that will inject or remove the trace
10482 kvm_set_rflags(vcpu, rflags);
10484 static_call(kvm_x86_update_exception_bitmap)(vcpu);
10494 * Translate a guest virtual address to a guest physical address.
10496 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
10497 struct kvm_translation *tr)
10499 unsigned long vaddr = tr->linear_address;
10505 idx = srcu_read_lock(&vcpu->kvm->srcu);
10506 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
10507 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10508 tr->physical_address = gpa;
10509 tr->valid = gpa != UNMAPPED_GVA;
10517 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10519 struct fxregs_state *fxsave;
10521 if (!vcpu->arch.guest_fpu)
10526 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10527 memcpy(fpu->fpr, fxsave->st_space, 128);
10528 fpu->fcw = fxsave->cwd;
10529 fpu->fsw = fxsave->swd;
10530 fpu->ftwx = fxsave->twd;
10531 fpu->last_opcode = fxsave->fop;
10532 fpu->last_ip = fxsave->rip;
10533 fpu->last_dp = fxsave->rdp;
10534 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
10540 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10542 struct fxregs_state *fxsave;
10544 if (!vcpu->arch.guest_fpu)
10549 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10551 memcpy(fxsave->st_space, fpu->fpr, 128);
10552 fxsave->cwd = fpu->fcw;
10553 fxsave->swd = fpu->fsw;
10554 fxsave->twd = fpu->ftwx;
10555 fxsave->fop = fpu->last_opcode;
10556 fxsave->rip = fpu->last_ip;
10557 fxsave->rdp = fpu->last_dp;
10558 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
10564 static void store_regs(struct kvm_vcpu *vcpu)
10566 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
10568 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
10569 __get_regs(vcpu, &vcpu->run->s.regs.regs);
10571 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
10572 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
10574 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
10575 kvm_vcpu_ioctl_x86_get_vcpu_events(
10576 vcpu, &vcpu->run->s.regs.events);
10579 static int sync_regs(struct kvm_vcpu *vcpu)
10581 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
10582 __set_regs(vcpu, &vcpu->run->s.regs.regs);
10583 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
10585 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
10586 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
10588 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
10590 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
10591 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
10592 vcpu, &vcpu->run->s.regs.events))
10594 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
10600 static void fx_init(struct kvm_vcpu *vcpu)
10602 if (!vcpu->arch.guest_fpu)
10605 fpstate_init(&vcpu->arch.guest_fpu->state);
10606 if (boot_cpu_has(X86_FEATURE_XSAVES))
10607 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
10608 host_xcr0 | XSTATE_COMPACTION_ENABLED;
10611 * Ensure guest xcr0 is valid for loading
10613 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10616 void kvm_free_guest_fpu(struct kvm_vcpu *vcpu)
10618 if (vcpu->arch.guest_fpu) {
10619 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
10620 vcpu->arch.guest_fpu = NULL;
10623 EXPORT_SYMBOL_GPL(kvm_free_guest_fpu);
10625 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10627 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10628 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10629 "guest TSC will not be reliable\n");
10634 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
10639 vcpu->arch.last_vmentry_cpu = -1;
10640 vcpu->arch.regs_avail = ~0;
10641 vcpu->arch.regs_dirty = ~0;
10643 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
10644 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10646 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
10648 r = kvm_mmu_create(vcpu);
10652 if (irqchip_in_kernel(vcpu->kvm)) {
10653 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
10655 goto fail_mmu_destroy;
10656 if (kvm_apicv_activated(vcpu->kvm))
10657 vcpu->arch.apicv_active = true;
10659 static_branch_inc(&kvm_has_noapic_vcpu);
10663 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
10665 goto fail_free_lapic;
10666 vcpu->arch.pio_data = page_address(page);
10668 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
10669 GFP_KERNEL_ACCOUNT);
10670 if (!vcpu->arch.mce_banks)
10671 goto fail_free_pio_data;
10672 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
10674 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
10675 GFP_KERNEL_ACCOUNT))
10676 goto fail_free_mce_banks;
10678 if (!alloc_emulate_ctxt(vcpu))
10679 goto free_wbinvd_dirty_mask;
10681 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
10682 GFP_KERNEL_ACCOUNT);
10683 if (!vcpu->arch.user_fpu) {
10684 pr_err("kvm: failed to allocate userspace's fpu\n");
10685 goto free_emulate_ctxt;
10688 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
10689 GFP_KERNEL_ACCOUNT);
10690 if (!vcpu->arch.guest_fpu) {
10691 pr_err("kvm: failed to allocate vcpu's fpu\n");
10692 goto free_user_fpu;
10696 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
10697 vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
10699 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10701 kvm_async_pf_hash_reset(vcpu);
10702 kvm_pmu_init(vcpu);
10704 vcpu->arch.pending_external_vector = -1;
10705 vcpu->arch.preempted_in_kernel = false;
10707 #if IS_ENABLED(CONFIG_HYPERV)
10708 vcpu->arch.hv_root_tdp = INVALID_PAGE;
10711 r = static_call(kvm_x86_vcpu_create)(vcpu);
10713 goto free_guest_fpu;
10715 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
10716 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
10717 kvm_vcpu_mtrr_init(vcpu);
10719 kvm_set_tsc_khz(vcpu, max_tsc_khz);
10720 kvm_vcpu_reset(vcpu, false);
10721 kvm_init_mmu(vcpu);
10726 kvm_free_guest_fpu(vcpu);
10728 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10730 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10731 free_wbinvd_dirty_mask:
10732 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10733 fail_free_mce_banks:
10734 kfree(vcpu->arch.mce_banks);
10735 fail_free_pio_data:
10736 free_page((unsigned long)vcpu->arch.pio_data);
10738 kvm_free_lapic(vcpu);
10740 kvm_mmu_destroy(vcpu);
10744 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
10746 struct kvm *kvm = vcpu->kvm;
10748 if (mutex_lock_killable(&vcpu->mutex))
10751 kvm_synchronize_tsc(vcpu, 0);
10754 /* poll control enabled by default */
10755 vcpu->arch.msr_kvm_poll_control = 1;
10757 mutex_unlock(&vcpu->mutex);
10759 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
10760 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
10761 KVMCLOCK_SYNC_PERIOD);
10764 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
10766 struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
10769 kvm_release_pfn(cache->pfn, cache->dirty, cache);
10771 kvmclock_reset(vcpu);
10773 static_call(kvm_x86_vcpu_free)(vcpu);
10775 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10776 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10777 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10778 kvm_free_guest_fpu(vcpu);
10780 kvm_hv_vcpu_uninit(vcpu);
10781 kvm_pmu_destroy(vcpu);
10782 kfree(vcpu->arch.mce_banks);
10783 kvm_free_lapic(vcpu);
10784 idx = srcu_read_lock(&vcpu->kvm->srcu);
10785 kvm_mmu_destroy(vcpu);
10786 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10787 free_page((unsigned long)vcpu->arch.pio_data);
10788 kvfree(vcpu->arch.cpuid_entries);
10789 if (!lapic_in_kernel(vcpu))
10790 static_branch_dec(&kvm_has_noapic_vcpu);
10793 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
10795 unsigned long old_cr0 = kvm_read_cr0(vcpu);
10796 unsigned long new_cr0;
10799 kvm_lapic_reset(vcpu, init_event);
10801 vcpu->arch.hflags = 0;
10803 vcpu->arch.smi_pending = 0;
10804 vcpu->arch.smi_count = 0;
10805 atomic_set(&vcpu->arch.nmi_queued, 0);
10806 vcpu->arch.nmi_pending = 0;
10807 vcpu->arch.nmi_injected = false;
10808 kvm_clear_interrupt_queue(vcpu);
10809 kvm_clear_exception_queue(vcpu);
10811 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10812 kvm_update_dr0123(vcpu);
10813 vcpu->arch.dr6 = DR6_ACTIVE_LOW;
10814 vcpu->arch.dr7 = DR7_FIXED_1;
10815 kvm_update_dr7(vcpu);
10817 vcpu->arch.cr2 = 0;
10819 kvm_make_request(KVM_REQ_EVENT, vcpu);
10820 vcpu->arch.apf.msr_en_val = 0;
10821 vcpu->arch.apf.msr_int_val = 0;
10822 vcpu->arch.st.msr_val = 0;
10824 kvmclock_reset(vcpu);
10826 kvm_clear_async_pf_completion_queue(vcpu);
10827 kvm_async_pf_hash_reset(vcpu);
10828 vcpu->arch.apf.halted = false;
10830 if (vcpu->arch.guest_fpu && kvm_mpx_supported()) {
10831 void *mpx_state_buffer;
10834 * To avoid have the INIT path from kvm_apic_has_events() that be
10835 * called with loaded FPU and does not let userspace fix the state.
10838 kvm_put_guest_fpu(vcpu);
10839 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10841 if (mpx_state_buffer)
10842 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
10843 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10845 if (mpx_state_buffer)
10846 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
10848 kvm_load_guest_fpu(vcpu);
10852 kvm_pmu_reset(vcpu);
10853 vcpu->arch.smbase = 0x30000;
10855 vcpu->arch.msr_misc_features_enables = 0;
10857 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10860 /* All GPRs except RDX (handled below) are zeroed on RESET/INIT. */
10861 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
10862 kvm_register_mark_dirty(vcpu, VCPU_REGS_RSP);
10865 * Fall back to KVM's default Family/Model/Stepping of 0x600 (P6/Athlon)
10866 * if no CPUID match is found. Note, it's impossible to get a match at
10867 * RESET since KVM emulates RESET before exposing the vCPU to userspace,
10868 * i.e. it'simpossible for kvm_cpuid() to find a valid entry on RESET.
10869 * But, go through the motions in case that's ever remedied.
10872 if (!kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy, true))
10874 kvm_rdx_write(vcpu, eax);
10876 vcpu->arch.ia32_xss = 0;
10878 static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
10880 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
10881 kvm_rip_write(vcpu, 0xfff0);
10883 vcpu->arch.cr3 = 0;
10884 kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
10887 * CR0.CD/NW are set on RESET, preserved on INIT. Note, some versions
10888 * of Intel's SDM list CD/NW as being set on INIT, but they contradict
10889 * (or qualify) that with a footnote stating that CD/NW are preserved.
10891 new_cr0 = X86_CR0_ET;
10893 new_cr0 |= (old_cr0 & (X86_CR0_NW | X86_CR0_CD));
10895 new_cr0 |= X86_CR0_NW | X86_CR0_CD;
10897 static_call(kvm_x86_set_cr0)(vcpu, new_cr0);
10898 static_call(kvm_x86_set_cr4)(vcpu, 0);
10899 static_call(kvm_x86_set_efer)(vcpu, 0);
10900 static_call(kvm_x86_update_exception_bitmap)(vcpu);
10903 * Reset the MMU context if paging was enabled prior to INIT (which is
10904 * implied if CR0.PG=1 as CR0 will be '0' prior to RESET). Unlike the
10905 * standard CR0/CR4/EFER modification paths, only CR0.PG needs to be
10906 * checked because it is unconditionally cleared on INIT and all other
10907 * paging related bits are ignored if paging is disabled, i.e. CR0.WP,
10908 * CR4, and EFER changes are all irrelevant if CR0.PG was '0'.
10910 if (old_cr0 & X86_CR0_PG)
10911 kvm_mmu_reset_context(vcpu);
10914 * Intel's SDM states that all TLB entries are flushed on INIT. AMD's
10915 * APM states the TLBs are untouched by INIT, but it also states that
10916 * the TLBs are flushed on "External initialization of the processor."
10917 * Flush the guest TLB regardless of vendor, there is no meaningful
10918 * benefit in relying on the guest to flush the TLB immediately after
10919 * INIT. A spurious TLB flush is benign and likely negligible from a
10920 * performance perspective.
10923 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
10925 EXPORT_SYMBOL_GPL(kvm_vcpu_reset);
10927 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
10929 struct kvm_segment cs;
10931 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10932 cs.selector = vector << 8;
10933 cs.base = vector << 12;
10934 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
10935 kvm_rip_write(vcpu, 0);
10937 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
10939 int kvm_arch_hardware_enable(void)
10942 struct kvm_vcpu *vcpu;
10947 bool stable, backwards_tsc = false;
10949 kvm_user_return_msr_cpu_online();
10950 ret = static_call(kvm_x86_hardware_enable)();
10954 local_tsc = rdtsc();
10955 stable = !kvm_check_tsc_unstable();
10956 list_for_each_entry(kvm, &vm_list, vm_list) {
10957 kvm_for_each_vcpu(i, vcpu, kvm) {
10958 if (!stable && vcpu->cpu == smp_processor_id())
10959 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10960 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
10961 backwards_tsc = true;
10962 if (vcpu->arch.last_host_tsc > max_tsc)
10963 max_tsc = vcpu->arch.last_host_tsc;
10969 * Sometimes, even reliable TSCs go backwards. This happens on
10970 * platforms that reset TSC during suspend or hibernate actions, but
10971 * maintain synchronization. We must compensate. Fortunately, we can
10972 * detect that condition here, which happens early in CPU bringup,
10973 * before any KVM threads can be running. Unfortunately, we can't
10974 * bring the TSCs fully up to date with real time, as we aren't yet far
10975 * enough into CPU bringup that we know how much real time has actually
10976 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
10977 * variables that haven't been updated yet.
10979 * So we simply find the maximum observed TSC above, then record the
10980 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
10981 * the adjustment will be applied. Note that we accumulate
10982 * adjustments, in case multiple suspend cycles happen before some VCPU
10983 * gets a chance to run again. In the event that no KVM threads get a
10984 * chance to run, we will miss the entire elapsed period, as we'll have
10985 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
10986 * loose cycle time. This isn't too big a deal, since the loss will be
10987 * uniform across all VCPUs (not to mention the scenario is extremely
10988 * unlikely). It is possible that a second hibernate recovery happens
10989 * much faster than a first, causing the observed TSC here to be
10990 * smaller; this would require additional padding adjustment, which is
10991 * why we set last_host_tsc to the local tsc observed here.
10993 * N.B. - this code below runs only on platforms with reliable TSC,
10994 * as that is the only way backwards_tsc is set above. Also note
10995 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
10996 * have the same delta_cyc adjustment applied if backwards_tsc
10997 * is detected. Note further, this adjustment is only done once,
10998 * as we reset last_host_tsc on all VCPUs to stop this from being
10999 * called multiple times (one for each physical CPU bringup).
11001 * Platforms with unreliable TSCs don't have to deal with this, they
11002 * will be compensated by the logic in vcpu_load, which sets the TSC to
11003 * catchup mode. This will catchup all VCPUs to real time, but cannot
11004 * guarantee that they stay in perfect synchronization.
11006 if (backwards_tsc) {
11007 u64 delta_cyc = max_tsc - local_tsc;
11008 list_for_each_entry(kvm, &vm_list, vm_list) {
11009 kvm->arch.backwards_tsc_observed = true;
11010 kvm_for_each_vcpu(i, vcpu, kvm) {
11011 vcpu->arch.tsc_offset_adjustment += delta_cyc;
11012 vcpu->arch.last_host_tsc = local_tsc;
11013 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
11017 * We have to disable TSC offset matching.. if you were
11018 * booting a VM while issuing an S4 host suspend....
11019 * you may have some problem. Solving this issue is
11020 * left as an exercise to the reader.
11022 kvm->arch.last_tsc_nsec = 0;
11023 kvm->arch.last_tsc_write = 0;
11030 void kvm_arch_hardware_disable(void)
11032 static_call(kvm_x86_hardware_disable)();
11033 drop_user_return_notifiers();
11036 int kvm_arch_hardware_setup(void *opaque)
11038 struct kvm_x86_init_ops *ops = opaque;
11041 rdmsrl_safe(MSR_EFER, &host_efer);
11043 if (boot_cpu_has(X86_FEATURE_XSAVES))
11044 rdmsrl(MSR_IA32_XSS, host_xss);
11046 r = ops->hardware_setup();
11050 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
11051 kvm_ops_static_call_update();
11053 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
11056 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
11057 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
11058 #undef __kvm_cpu_cap_has
11060 if (kvm_has_tsc_control) {
11062 * Make sure the user can only configure tsc_khz values that
11063 * fit into a signed integer.
11064 * A min value is not calculated because it will always
11065 * be 1 on all machines.
11067 u64 max = min(0x7fffffffULL,
11068 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
11069 kvm_max_guest_tsc_khz = max;
11071 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
11074 kvm_init_msr_list();
11078 void kvm_arch_hardware_unsetup(void)
11080 static_call(kvm_x86_hardware_unsetup)();
11083 int kvm_arch_check_processor_compat(void *opaque)
11085 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
11086 struct kvm_x86_init_ops *ops = opaque;
11088 WARN_ON(!irqs_disabled());
11090 if (__cr4_reserved_bits(cpu_has, c) !=
11091 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
11094 return ops->check_processor_compatibility();
11097 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
11099 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
11101 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
11103 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
11105 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
11108 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
11109 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
11111 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
11113 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
11115 vcpu->arch.l1tf_flush_l1d = true;
11116 if (pmu->version && unlikely(pmu->event_count)) {
11117 pmu->need_cleanup = true;
11118 kvm_make_request(KVM_REQ_PMU, vcpu);
11120 static_call(kvm_x86_sched_in)(vcpu, cpu);
11123 void kvm_arch_free_vm(struct kvm *kvm)
11125 kfree(to_kvm_hv(kvm)->hv_pa_pg);
11130 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
11137 ret = kvm_page_track_init(kvm);
11141 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
11142 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
11143 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
11144 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
11145 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
11146 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
11148 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
11149 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
11150 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
11151 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
11152 &kvm->arch.irq_sources_bitmap);
11154 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
11155 mutex_init(&kvm->arch.apic_map_lock);
11156 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
11158 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
11159 pvclock_update_vm_gtod_copy(kvm);
11161 kvm->arch.guest_can_read_msr_platform_info = true;
11163 #if IS_ENABLED(CONFIG_HYPERV)
11164 spin_lock_init(&kvm->arch.hv_root_tdp_lock);
11165 kvm->arch.hv_root_tdp = INVALID_PAGE;
11168 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
11169 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
11171 kvm_apicv_init(kvm);
11172 kvm_hv_init_vm(kvm);
11173 kvm_mmu_init_vm(kvm);
11174 kvm_xen_init_vm(kvm);
11176 return static_call(kvm_x86_vm_init)(kvm);
11179 int kvm_arch_post_init_vm(struct kvm *kvm)
11181 return kvm_mmu_post_init_vm(kvm);
11184 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
11187 kvm_mmu_unload(vcpu);
11191 static void kvm_free_vcpus(struct kvm *kvm)
11194 struct kvm_vcpu *vcpu;
11197 * Unpin any mmu pages first.
11199 kvm_for_each_vcpu(i, vcpu, kvm) {
11200 kvm_clear_async_pf_completion_queue(vcpu);
11201 kvm_unload_vcpu_mmu(vcpu);
11203 kvm_for_each_vcpu(i, vcpu, kvm)
11204 kvm_vcpu_destroy(vcpu);
11206 mutex_lock(&kvm->lock);
11207 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
11208 kvm->vcpus[i] = NULL;
11210 atomic_set(&kvm->online_vcpus, 0);
11211 mutex_unlock(&kvm->lock);
11214 void kvm_arch_sync_events(struct kvm *kvm)
11216 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
11217 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
11221 #define ERR_PTR_USR(e) ((void __user *)ERR_PTR(e))
11224 * __x86_set_memory_region: Setup KVM internal memory slot
11226 * @kvm: the kvm pointer to the VM.
11227 * @id: the slot ID to setup.
11228 * @gpa: the GPA to install the slot (unused when @size == 0).
11229 * @size: the size of the slot. Set to zero to uninstall a slot.
11231 * This function helps to setup a KVM internal memory slot. Specify
11232 * @size > 0 to install a new slot, while @size == 0 to uninstall a
11233 * slot. The return code can be one of the following:
11235 * HVA: on success (uninstall will return a bogus HVA)
11238 * The caller should always use IS_ERR() to check the return value
11239 * before use. Note, the KVM internal memory slots are guaranteed to
11240 * remain valid and unchanged until the VM is destroyed, i.e., the
11241 * GPA->HVA translation will not change. However, the HVA is a user
11242 * address, i.e. its accessibility is not guaranteed, and must be
11243 * accessed via __copy_{to,from}_user().
11245 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
11249 unsigned long hva, old_npages;
11250 struct kvm_memslots *slots = kvm_memslots(kvm);
11251 struct kvm_memory_slot *slot;
11253 /* Called with kvm->slots_lock held. */
11254 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
11255 return ERR_PTR_USR(-EINVAL);
11257 slot = id_to_memslot(slots, id);
11259 if (slot && slot->npages)
11260 return ERR_PTR_USR(-EEXIST);
11263 * MAP_SHARED to prevent internal slot pages from being moved
11266 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
11267 MAP_SHARED | MAP_ANONYMOUS, 0);
11268 if (IS_ERR((void *)hva))
11269 return (void __user *)hva;
11271 if (!slot || !slot->npages)
11274 old_npages = slot->npages;
11275 hva = slot->userspace_addr;
11278 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11279 struct kvm_userspace_memory_region m;
11281 m.slot = id | (i << 16);
11283 m.guest_phys_addr = gpa;
11284 m.userspace_addr = hva;
11285 m.memory_size = size;
11286 r = __kvm_set_memory_region(kvm, &m);
11288 return ERR_PTR_USR(r);
11292 vm_munmap(hva, old_npages * PAGE_SIZE);
11294 return (void __user *)hva;
11296 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
11298 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
11300 kvm_mmu_pre_destroy_vm(kvm);
11303 void kvm_arch_destroy_vm(struct kvm *kvm)
11305 if (current->mm == kvm->mm) {
11307 * Free memory regions allocated on behalf of userspace,
11308 * unless the the memory map has changed due to process exit
11311 mutex_lock(&kvm->slots_lock);
11312 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
11314 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
11316 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
11317 mutex_unlock(&kvm->slots_lock);
11319 static_call_cond(kvm_x86_vm_destroy)(kvm);
11320 kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
11321 kvm_pic_destroy(kvm);
11322 kvm_ioapic_destroy(kvm);
11323 kvm_free_vcpus(kvm);
11324 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
11325 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
11326 kvm_mmu_uninit_vm(kvm);
11327 kvm_page_track_cleanup(kvm);
11328 kvm_xen_destroy_vm(kvm);
11329 kvm_hv_destroy_vm(kvm);
11332 static void memslot_rmap_free(struct kvm_memory_slot *slot)
11336 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11337 kvfree(slot->arch.rmap[i]);
11338 slot->arch.rmap[i] = NULL;
11342 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
11346 memslot_rmap_free(slot);
11348 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11349 kvfree(slot->arch.lpage_info[i - 1]);
11350 slot->arch.lpage_info[i - 1] = NULL;
11353 kvm_page_track_free_memslot(slot);
11356 static int memslot_rmap_alloc(struct kvm_memory_slot *slot,
11357 unsigned long npages)
11359 const int sz = sizeof(*slot->arch.rmap[0]);
11362 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11364 int lpages = __kvm_mmu_slot_lpages(slot, npages, level);
11366 WARN_ON(slot->arch.rmap[i]);
11368 slot->arch.rmap[i] = kvcalloc(lpages, sz, GFP_KERNEL_ACCOUNT);
11369 if (!slot->arch.rmap[i]) {
11370 memslot_rmap_free(slot);
11378 int alloc_all_memslots_rmaps(struct kvm *kvm)
11380 struct kvm_memslots *slots;
11381 struct kvm_memory_slot *slot;
11385 * Check if memslots alreday have rmaps early before acquiring
11386 * the slots_arch_lock below.
11388 if (kvm_memslots_have_rmaps(kvm))
11391 mutex_lock(&kvm->slots_arch_lock);
11394 * Read memslots_have_rmaps again, under the slots arch lock,
11395 * before allocating the rmaps
11397 if (kvm_memslots_have_rmaps(kvm)) {
11398 mutex_unlock(&kvm->slots_arch_lock);
11402 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11403 slots = __kvm_memslots(kvm, i);
11404 kvm_for_each_memslot(slot, slots) {
11405 r = memslot_rmap_alloc(slot, slot->npages);
11407 mutex_unlock(&kvm->slots_arch_lock);
11414 * Ensure that memslots_have_rmaps becomes true strictly after
11415 * all the rmap pointers are set.
11417 smp_store_release(&kvm->arch.memslots_have_rmaps, true);
11418 mutex_unlock(&kvm->slots_arch_lock);
11422 static int kvm_alloc_memslot_metadata(struct kvm *kvm,
11423 struct kvm_memory_slot *slot,
11424 unsigned long npages)
11429 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
11430 * old arrays will be freed by __kvm_set_memory_region() if installing
11431 * the new memslot is successful.
11433 memset(&slot->arch, 0, sizeof(slot->arch));
11435 if (kvm_memslots_have_rmaps(kvm)) {
11436 r = memslot_rmap_alloc(slot, npages);
11441 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11442 struct kvm_lpage_info *linfo;
11443 unsigned long ugfn;
11447 lpages = __kvm_mmu_slot_lpages(slot, npages, level);
11449 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
11453 slot->arch.lpage_info[i - 1] = linfo;
11455 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
11456 linfo[0].disallow_lpage = 1;
11457 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
11458 linfo[lpages - 1].disallow_lpage = 1;
11459 ugfn = slot->userspace_addr >> PAGE_SHIFT;
11461 * If the gfn and userspace address are not aligned wrt each
11462 * other, disable large page support for this slot.
11464 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
11467 for (j = 0; j < lpages; ++j)
11468 linfo[j].disallow_lpage = 1;
11472 if (kvm_page_track_create_memslot(slot, npages))
11478 memslot_rmap_free(slot);
11480 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11481 kvfree(slot->arch.lpage_info[i - 1]);
11482 slot->arch.lpage_info[i - 1] = NULL;
11487 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
11489 struct kvm_vcpu *vcpu;
11493 * memslots->generation has been incremented.
11494 * mmio generation may have reached its maximum value.
11496 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
11498 /* Force re-initialization of steal_time cache */
11499 kvm_for_each_vcpu(i, vcpu, kvm)
11500 kvm_vcpu_kick(vcpu);
11503 int kvm_arch_prepare_memory_region(struct kvm *kvm,
11504 struct kvm_memory_slot *memslot,
11505 const struct kvm_userspace_memory_region *mem,
11506 enum kvm_mr_change change)
11508 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
11509 return kvm_alloc_memslot_metadata(kvm, memslot,
11510 mem->memory_size >> PAGE_SHIFT);
11515 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
11517 struct kvm_arch *ka = &kvm->arch;
11519 if (!kvm_x86_ops.cpu_dirty_log_size)
11522 if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
11523 (!enable && --ka->cpu_dirty_logging_count == 0))
11524 kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
11526 WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
11529 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
11530 struct kvm_memory_slot *old,
11531 const struct kvm_memory_slot *new,
11532 enum kvm_mr_change change)
11534 bool log_dirty_pages = new->flags & KVM_MEM_LOG_DIRTY_PAGES;
11537 * Update CPU dirty logging if dirty logging is being toggled. This
11538 * applies to all operations.
11540 if ((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)
11541 kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
11544 * Nothing more to do for RO slots (which can't be dirtied and can't be
11545 * made writable) or CREATE/MOVE/DELETE of a slot.
11547 * For a memslot with dirty logging disabled:
11548 * CREATE: No dirty mappings will already exist.
11549 * MOVE/DELETE: The old mappings will already have been cleaned up by
11550 * kvm_arch_flush_shadow_memslot()
11552 * For a memslot with dirty logging enabled:
11553 * CREATE: No shadow pages exist, thus nothing to write-protect
11554 * and no dirty bits to clear.
11555 * MOVE/DELETE: The old mappings will already have been cleaned up by
11556 * kvm_arch_flush_shadow_memslot().
11558 if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
11562 * READONLY and non-flags changes were filtered out above, and the only
11563 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
11564 * logging isn't being toggled on or off.
11566 if (WARN_ON_ONCE(!((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)))
11569 if (!log_dirty_pages) {
11571 * Dirty logging tracks sptes in 4k granularity, meaning that
11572 * large sptes have to be split. If live migration succeeds,
11573 * the guest in the source machine will be destroyed and large
11574 * sptes will be created in the destination. However, if the
11575 * guest continues to run in the source machine (for example if
11576 * live migration fails), small sptes will remain around and
11577 * cause bad performance.
11579 * Scan sptes if dirty logging has been stopped, dropping those
11580 * which can be collapsed into a single large-page spte. Later
11581 * page faults will create the large-page sptes.
11583 kvm_mmu_zap_collapsible_sptes(kvm, new);
11586 * Initially-all-set does not require write protecting any page,
11587 * because they're all assumed to be dirty.
11589 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
11592 if (kvm_x86_ops.cpu_dirty_log_size) {
11593 kvm_mmu_slot_leaf_clear_dirty(kvm, new);
11594 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_2M);
11596 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_4K);
11601 void kvm_arch_commit_memory_region(struct kvm *kvm,
11602 const struct kvm_userspace_memory_region *mem,
11603 struct kvm_memory_slot *old,
11604 const struct kvm_memory_slot *new,
11605 enum kvm_mr_change change)
11607 if (!kvm->arch.n_requested_mmu_pages)
11608 kvm_mmu_change_mmu_pages(kvm,
11609 kvm_mmu_calculate_default_mmu_pages(kvm));
11611 kvm_mmu_slot_apply_flags(kvm, old, new, change);
11613 /* Free the arrays associated with the old memslot. */
11614 if (change == KVM_MR_MOVE)
11615 kvm_arch_free_memslot(kvm, old);
11618 void kvm_arch_flush_shadow_all(struct kvm *kvm)
11620 kvm_mmu_zap_all(kvm);
11623 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
11624 struct kvm_memory_slot *slot)
11626 kvm_page_track_flush_slot(kvm, slot);
11629 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
11631 return (is_guest_mode(vcpu) &&
11632 kvm_x86_ops.guest_apic_has_interrupt &&
11633 static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
11636 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
11638 if (!list_empty_careful(&vcpu->async_pf.done))
11641 if (kvm_apic_has_events(vcpu))
11644 if (vcpu->arch.pv.pv_unhalted)
11647 if (vcpu->arch.exception.pending)
11650 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11651 (vcpu->arch.nmi_pending &&
11652 static_call(kvm_x86_nmi_allowed)(vcpu, false)))
11655 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
11656 (vcpu->arch.smi_pending &&
11657 static_call(kvm_x86_smi_allowed)(vcpu, false)))
11660 if (kvm_arch_interrupt_allowed(vcpu) &&
11661 (kvm_cpu_has_interrupt(vcpu) ||
11662 kvm_guest_apic_has_interrupt(vcpu)))
11665 if (kvm_hv_has_stimer_pending(vcpu))
11668 if (is_guest_mode(vcpu) &&
11669 kvm_x86_ops.nested_ops->hv_timer_pending &&
11670 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
11676 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
11678 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
11681 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
11683 if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
11689 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
11691 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
11694 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11695 kvm_test_request(KVM_REQ_SMI, vcpu) ||
11696 kvm_test_request(KVM_REQ_EVENT, vcpu))
11699 return kvm_arch_dy_has_pending_interrupt(vcpu);
11702 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
11704 if (vcpu->arch.guest_state_protected)
11707 return vcpu->arch.preempted_in_kernel;
11710 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
11712 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
11715 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
11717 return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
11720 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
11722 /* Can't read the RIP when guest state is protected, just return 0 */
11723 if (vcpu->arch.guest_state_protected)
11726 if (is_64_bit_mode(vcpu))
11727 return kvm_rip_read(vcpu);
11728 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
11729 kvm_rip_read(vcpu));
11731 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
11733 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
11735 return kvm_get_linear_rip(vcpu) == linear_rip;
11737 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
11739 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
11741 unsigned long rflags;
11743 rflags = static_call(kvm_x86_get_rflags)(vcpu);
11744 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11745 rflags &= ~X86_EFLAGS_TF;
11748 EXPORT_SYMBOL_GPL(kvm_get_rflags);
11750 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11752 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
11753 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
11754 rflags |= X86_EFLAGS_TF;
11755 static_call(kvm_x86_set_rflags)(vcpu, rflags);
11758 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11760 __kvm_set_rflags(vcpu, rflags);
11761 kvm_make_request(KVM_REQ_EVENT, vcpu);
11763 EXPORT_SYMBOL_GPL(kvm_set_rflags);
11765 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
11769 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
11773 r = kvm_mmu_reload(vcpu);
11777 if (!vcpu->arch.mmu->direct_map &&
11778 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
11781 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
11784 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
11786 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
11788 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
11791 static inline u32 kvm_async_pf_next_probe(u32 key)
11793 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
11796 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11798 u32 key = kvm_async_pf_hash_fn(gfn);
11800 while (vcpu->arch.apf.gfns[key] != ~0)
11801 key = kvm_async_pf_next_probe(key);
11803 vcpu->arch.apf.gfns[key] = gfn;
11806 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
11809 u32 key = kvm_async_pf_hash_fn(gfn);
11811 for (i = 0; i < ASYNC_PF_PER_VCPU &&
11812 (vcpu->arch.apf.gfns[key] != gfn &&
11813 vcpu->arch.apf.gfns[key] != ~0); i++)
11814 key = kvm_async_pf_next_probe(key);
11819 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11821 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
11824 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11828 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
11830 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
11834 vcpu->arch.apf.gfns[i] = ~0;
11836 j = kvm_async_pf_next_probe(j);
11837 if (vcpu->arch.apf.gfns[j] == ~0)
11839 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
11841 * k lies cyclically in ]i,j]
11843 * |....j i.k.| or |.k..j i...|
11845 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
11846 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
11851 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
11853 u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
11855 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
11859 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
11861 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11863 return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11864 &token, offset, sizeof(token));
11867 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
11869 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11872 if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11873 &val, offset, sizeof(val)))
11879 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
11881 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
11884 if (!kvm_pv_async_pf_enabled(vcpu) ||
11885 (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
11891 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
11893 if (unlikely(!lapic_in_kernel(vcpu) ||
11894 kvm_event_needs_reinjection(vcpu) ||
11895 vcpu->arch.exception.pending))
11898 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
11902 * If interrupts are off we cannot even use an artificial
11905 return kvm_arch_interrupt_allowed(vcpu);
11908 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
11909 struct kvm_async_pf *work)
11911 struct x86_exception fault;
11913 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
11914 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
11916 if (kvm_can_deliver_async_pf(vcpu) &&
11917 !apf_put_user_notpresent(vcpu)) {
11918 fault.vector = PF_VECTOR;
11919 fault.error_code_valid = true;
11920 fault.error_code = 0;
11921 fault.nested_page_fault = false;
11922 fault.address = work->arch.token;
11923 fault.async_page_fault = true;
11924 kvm_inject_page_fault(vcpu, &fault);
11928 * It is not possible to deliver a paravirtualized asynchronous
11929 * page fault, but putting the guest in an artificial halt state
11930 * can be beneficial nevertheless: if an interrupt arrives, we
11931 * can deliver it timely and perhaps the guest will schedule
11932 * another process. When the instruction that triggered a page
11933 * fault is retried, hopefully the page will be ready in the host.
11935 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
11940 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
11941 struct kvm_async_pf *work)
11943 struct kvm_lapic_irq irq = {
11944 .delivery_mode = APIC_DM_FIXED,
11945 .vector = vcpu->arch.apf.vec
11948 if (work->wakeup_all)
11949 work->arch.token = ~0; /* broadcast wakeup */
11951 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
11952 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
11954 if ((work->wakeup_all || work->notpresent_injected) &&
11955 kvm_pv_async_pf_enabled(vcpu) &&
11956 !apf_put_user_ready(vcpu, work->arch.token)) {
11957 vcpu->arch.apf.pageready_pending = true;
11958 kvm_apic_set_irq(vcpu, &irq, NULL);
11961 vcpu->arch.apf.halted = false;
11962 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11965 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
11967 kvm_make_request(KVM_REQ_APF_READY, vcpu);
11968 if (!vcpu->arch.apf.pageready_pending)
11969 kvm_vcpu_kick(vcpu);
11972 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
11974 if (!kvm_pv_async_pf_enabled(vcpu))
11977 return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
11980 void kvm_arch_start_assignment(struct kvm *kvm)
11982 if (atomic_inc_return(&kvm->arch.assigned_device_count) == 1)
11983 static_call_cond(kvm_x86_start_assignment)(kvm);
11985 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
11987 void kvm_arch_end_assignment(struct kvm *kvm)
11989 atomic_dec(&kvm->arch.assigned_device_count);
11991 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
11993 bool kvm_arch_has_assigned_device(struct kvm *kvm)
11995 return atomic_read(&kvm->arch.assigned_device_count);
11997 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
11999 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
12001 atomic_inc(&kvm->arch.noncoherent_dma_count);
12003 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
12005 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
12007 atomic_dec(&kvm->arch.noncoherent_dma_count);
12009 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
12011 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
12013 return atomic_read(&kvm->arch.noncoherent_dma_count);
12015 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
12017 bool kvm_arch_has_irq_bypass(void)
12022 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
12023 struct irq_bypass_producer *prod)
12025 struct kvm_kernel_irqfd *irqfd =
12026 container_of(cons, struct kvm_kernel_irqfd, consumer);
12029 irqfd->producer = prod;
12030 kvm_arch_start_assignment(irqfd->kvm);
12031 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
12032 prod->irq, irqfd->gsi, 1);
12035 kvm_arch_end_assignment(irqfd->kvm);
12040 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
12041 struct irq_bypass_producer *prod)
12044 struct kvm_kernel_irqfd *irqfd =
12045 container_of(cons, struct kvm_kernel_irqfd, consumer);
12047 WARN_ON(irqfd->producer != prod);
12048 irqfd->producer = NULL;
12051 * When producer of consumer is unregistered, we change back to
12052 * remapped mode, so we can re-use the current implementation
12053 * when the irq is masked/disabled or the consumer side (KVM
12054 * int this case doesn't want to receive the interrupts.
12056 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
12058 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
12059 " fails: %d\n", irqfd->consumer.token, ret);
12061 kvm_arch_end_assignment(irqfd->kvm);
12064 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
12065 uint32_t guest_irq, bool set)
12067 return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
12070 bool kvm_vector_hashing_enabled(void)
12072 return vector_hashing;
12075 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
12077 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
12079 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
12082 int kvm_spec_ctrl_test_value(u64 value)
12085 * test that setting IA32_SPEC_CTRL to given value
12086 * is allowed by the host processor
12090 unsigned long flags;
12093 local_irq_save(flags);
12095 if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
12097 else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
12100 wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
12102 local_irq_restore(flags);
12106 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
12108 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
12110 struct x86_exception fault;
12111 u32 access = error_code &
12112 (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
12114 if (!(error_code & PFERR_PRESENT_MASK) ||
12115 vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
12117 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
12118 * tables probably do not match the TLB. Just proceed
12119 * with the error code that the processor gave.
12121 fault.vector = PF_VECTOR;
12122 fault.error_code_valid = true;
12123 fault.error_code = error_code;
12124 fault.nested_page_fault = false;
12125 fault.address = gva;
12127 vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
12129 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
12132 * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
12133 * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
12134 * indicates whether exit to userspace is needed.
12136 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
12137 struct x86_exception *e)
12139 if (r == X86EMUL_PROPAGATE_FAULT) {
12140 kvm_inject_emulated_page_fault(vcpu, e);
12145 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
12146 * while handling a VMX instruction KVM could've handled the request
12147 * correctly by exiting to userspace and performing I/O but there
12148 * doesn't seem to be a real use-case behind such requests, just return
12149 * KVM_EXIT_INTERNAL_ERROR for now.
12151 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
12152 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
12153 vcpu->run->internal.ndata = 0;
12157 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
12159 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
12162 struct x86_exception e;
12169 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
12170 if (r != X86EMUL_CONTINUE)
12171 return kvm_handle_memory_failure(vcpu, r, &e);
12173 if (operand.pcid >> 12 != 0) {
12174 kvm_inject_gp(vcpu, 0);
12178 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
12181 case INVPCID_TYPE_INDIV_ADDR:
12182 if ((!pcid_enabled && (operand.pcid != 0)) ||
12183 is_noncanonical_address(operand.gla, vcpu)) {
12184 kvm_inject_gp(vcpu, 0);
12187 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
12188 return kvm_skip_emulated_instruction(vcpu);
12190 case INVPCID_TYPE_SINGLE_CTXT:
12191 if (!pcid_enabled && (operand.pcid != 0)) {
12192 kvm_inject_gp(vcpu, 0);
12196 kvm_invalidate_pcid(vcpu, operand.pcid);
12197 return kvm_skip_emulated_instruction(vcpu);
12199 case INVPCID_TYPE_ALL_NON_GLOBAL:
12201 * Currently, KVM doesn't mark global entries in the shadow
12202 * page tables, so a non-global flush just degenerates to a
12203 * global flush. If needed, we could optimize this later by
12204 * keeping track of global entries in shadow page tables.
12208 case INVPCID_TYPE_ALL_INCL_GLOBAL:
12209 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
12210 return kvm_skip_emulated_instruction(vcpu);
12213 BUG(); /* We have already checked above that type <= 3 */
12216 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
12218 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
12220 struct kvm_run *run = vcpu->run;
12221 struct kvm_mmio_fragment *frag;
12224 BUG_ON(!vcpu->mmio_needed);
12226 /* Complete previous fragment */
12227 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
12228 len = min(8u, frag->len);
12229 if (!vcpu->mmio_is_write)
12230 memcpy(frag->data, run->mmio.data, len);
12232 if (frag->len <= 8) {
12233 /* Switch to the next fragment. */
12235 vcpu->mmio_cur_fragment++;
12237 /* Go forward to the next mmio piece. */
12243 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
12244 vcpu->mmio_needed = 0;
12246 // VMG change, at this point, we're always done
12247 // RIP has already been advanced
12251 // More MMIO is needed
12252 run->mmio.phys_addr = frag->gpa;
12253 run->mmio.len = min(8u, frag->len);
12254 run->mmio.is_write = vcpu->mmio_is_write;
12255 if (run->mmio.is_write)
12256 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
12257 run->exit_reason = KVM_EXIT_MMIO;
12259 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12264 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12268 struct kvm_mmio_fragment *frag;
12273 handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12274 if (handled == bytes)
12281 /*TODO: Check if need to increment number of frags */
12282 frag = vcpu->mmio_fragments;
12283 vcpu->mmio_nr_fragments = 1;
12288 vcpu->mmio_needed = 1;
12289 vcpu->mmio_cur_fragment = 0;
12291 vcpu->run->mmio.phys_addr = gpa;
12292 vcpu->run->mmio.len = min(8u, frag->len);
12293 vcpu->run->mmio.is_write = 1;
12294 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
12295 vcpu->run->exit_reason = KVM_EXIT_MMIO;
12297 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12301 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
12303 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12307 struct kvm_mmio_fragment *frag;
12312 handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12313 if (handled == bytes)
12320 /*TODO: Check if need to increment number of frags */
12321 frag = vcpu->mmio_fragments;
12322 vcpu->mmio_nr_fragments = 1;
12327 vcpu->mmio_needed = 1;
12328 vcpu->mmio_cur_fragment = 0;
12330 vcpu->run->mmio.phys_addr = gpa;
12331 vcpu->run->mmio.len = min(8u, frag->len);
12332 vcpu->run->mmio.is_write = 0;
12333 vcpu->run->exit_reason = KVM_EXIT_MMIO;
12335 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12339 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
12341 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
12343 memcpy(vcpu->arch.guest_ins_data, vcpu->arch.pio_data,
12344 vcpu->arch.pio.count * vcpu->arch.pio.size);
12345 vcpu->arch.pio.count = 0;
12350 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12351 unsigned int port, void *data, unsigned int count)
12355 ret = emulator_pio_out_emulated(vcpu->arch.emulate_ctxt, size, port,
12360 vcpu->arch.pio.count = 0;
12365 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
12366 unsigned int port, void *data, unsigned int count)
12370 ret = emulator_pio_in_emulated(vcpu->arch.emulate_ctxt, size, port,
12373 vcpu->arch.pio.count = 0;
12375 vcpu->arch.guest_ins_data = data;
12376 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
12382 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
12383 unsigned int port, void *data, unsigned int count,
12386 return in ? kvm_sev_es_ins(vcpu, size, port, data, count)
12387 : kvm_sev_es_outs(vcpu, size, port, data, count);
12389 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
12391 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
12392 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
12393 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
12394 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
12395 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
12396 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
12397 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
12398 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
12399 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
12400 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
12401 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
12402 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
12403 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
12404 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
12405 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
12406 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
12407 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
12408 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
12409 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
12410 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
12411 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
12412 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
12413 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
12414 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
12415 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
12416 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
12417 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);